The morphological productivity of selected ... - Helda - Helsinki.fi
The morphological productivity of selected ... - Helda - Helsinki.fi
The morphological productivity of selected ... - Helda - Helsinki.fi
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>The</strong> <strong>morphological</strong> <strong>productivity</strong> <strong>of</strong> <strong>selected</strong> combining forms in<br />
English<br />
Pro gradu thesis<br />
Department <strong>of</strong> modern languages<br />
University <strong>of</strong> <strong>Helsinki</strong><br />
26 September 2013<br />
Eeva Rita-Kasari
Contents<br />
1. Introduction ................................................................................................................ 1<br />
2. Background ................................................................................................................ 4<br />
2.1. Word-formation ....................................................................................................... 4<br />
2.1.1. Basic concepts <strong>of</strong> word-formation ....................................................................... 4<br />
2.1.2. <strong>The</strong> basic notions <strong>of</strong> the English word-formation system .................................... 6<br />
2.1.2. <strong>The</strong> diachronic perspective on the English word-formation ................................ 8<br />
2.2. Combining forms .................................................................................................. 10<br />
2.2.1 De<strong>fi</strong>ning combining forms .................................................................................. 10<br />
2.2.2. Neoclassical compounds .................................................................................... 13<br />
2.2.3. <strong>The</strong> status <strong>of</strong> combining forms ........................................................................... 15<br />
2.3. Morphological <strong>productivity</strong> .................................................................................. 18<br />
2.3.1. De<strong>fi</strong>ning <strong>morphological</strong> <strong>productivity</strong> ................................................................. 18<br />
2.3.2. Qualitative versus quantitative approaches to <strong>productivity</strong> ............................... 22<br />
2.3.3. Measuring <strong>morphological</strong> <strong>productivity</strong> .............................................................. 23<br />
2.3.3. Lexical statistics and <strong>productivity</strong> ...................................................................... 30<br />
2.3.4. <strong>The</strong> psycholinguistic aspect <strong>of</strong> <strong>productivity</strong> ...................................................... 34<br />
2.3.4. Constraints on <strong>productivity</strong> ................................................................................ 35<br />
3. Material and methods ............................................................................................... 39<br />
3.1. A closer look at the <strong>selected</strong> combining forms ..................................................... 39<br />
3.2. Corpora versus dictionaries in the study <strong>of</strong> <strong>productivity</strong> ...................................... 42<br />
3.3. How to determine units <strong>of</strong> analysis and other methodological problems ............. 45<br />
4. Results ...................................................................................................................... 49<br />
4.1. Potential <strong>productivity</strong> P ......................................................................................... 49<br />
4.2. LNRE modelling ................................................................................................... 54<br />
5. Discussion ................................................................................................................ 60<br />
6. Conclusion ................................................................................................................ 65
Primary sources and s<strong>of</strong>tware ...................................................................................... 68<br />
References .................................................................................................................... 68
1. Introduction<br />
Morphological <strong>productivity</strong>, the ability <strong>of</strong> speakers to coin new words using the<br />
resources <strong>of</strong> a given language, has received a great deal <strong>of</strong> attention in the past<br />
few decades, especially in the context <strong>of</strong> the English language. Bauer (2001) and<br />
Plag (1999), for example, <strong>of</strong>fer book-length accounts <strong>of</strong> this phenomenon.<br />
According to Bauer, “<strong>productivity</strong> remains one <strong>of</strong> the most contested areas in the<br />
study <strong>of</strong> word-formation” (1983: 62). Even though Bauer’s book was published<br />
almost thirty years ago and many new theories on <strong>morphological</strong> <strong>productivity</strong><br />
have been proposed since then, this statement still seems to be valid. Even though<br />
the concept <strong>of</strong> <strong>productivity</strong> seems to be easy to grasp intuitively, there are several<br />
practical and theoretical issues that have not yet been tackled.<br />
Baayen and his collaborators in particular have developed a number<br />
<strong>of</strong> empirical corpus-based methods to gauge <strong>productivity</strong> quantitatively (see, e.g.,<br />
Baayen 1992, 1993, 2009; Baayen and Lieber 1991; Chitashvili and Baayen 1993;<br />
Baayen and Renouf 1996). In addition, the psycholinguistic aspect <strong>of</strong><br />
<strong>productivity</strong>, i.e., the way speakers store complex words in their mental lexicon,<br />
has received considerable attention since the early nineties. Important<br />
contributions have come from Frauenfelder and Schreuder (1992), Baayen (1993),<br />
and Hay and Baayen (2002).<br />
In addition to being an important theoretical concept, studying<br />
<strong>morphological</strong> <strong>productivity</strong> also has practical relevance. Measuring degrees <strong>of</strong><br />
<strong>productivity</strong> is important from the point <strong>of</strong> view <strong>of</strong> writing grammars, for<br />
example. While unproductive processes are listed in the lexicon, productive<br />
patterns can be described by rules, and thus measuring <strong>productivity</strong> <strong>of</strong>fers<br />
valuable information on what processes should be given prominence in the case <strong>of</strong><br />
limited resources (Lüdeling et al. 2000: 57). <strong>The</strong> importance <strong>of</strong> the concept <strong>of</strong><br />
<strong>productivity</strong> can also be seen in the fact that computational tools cannot function<br />
properly unless they take productive word-formation into account (Baayen 2008:<br />
900). In addition, knowledge <strong>of</strong> the <strong>productivity</strong> <strong>of</strong> different word-formation<br />
processes may also be relevant in language teaching. Since combining forms are<br />
particularly frequent in scienti<strong>fi</strong>c registers, this study might also have some<br />
1
elevance in the context <strong>of</strong> English for special and academic purposes (see Fradin<br />
2000: 22–23).<br />
Studies concentrating on the <strong>productivity</strong> <strong>of</strong> English af<strong>fi</strong>xes abound.<br />
Most <strong>of</strong> them have focused on suf<strong>fi</strong>xes (see, e.g., Baayen and Lieber 1991;<br />
Baayen and Renouf 1996; Bauer 1992; Plag et al. 1999; Plag 2006, Palmer 2009,<br />
Säily and Suomela 2009, Säily 2011). This might be due to several reasons, one <strong>of</strong><br />
which being their class-changing nature. Pre<strong>fi</strong>xation, on the other hand, has not<br />
received the same amount <strong>of</strong> attention. Plag writes that in spite <strong>of</strong> the fact that<br />
English is one <strong>of</strong> the best-described language in the world, “the exact properties <strong>of</strong><br />
many af<strong>fi</strong>xes are still not suf<strong>fi</strong>ciently well determined” (2003: 86), and that there<br />
is still a great need for more detailed investigation.<br />
Neoclassical compounding and what the Oxford English Dictionary<br />
calls combining forms have been studied even less, and the few exceptions (e.g.,<br />
Warren 1990, Prćić 2005, Prćić 2007, Prćić 2008, McCauley 2006, and Kastovsky<br />
2009) have mainly been rather theoretical, and nothing really explicit has been<br />
said about their <strong>productivity</strong>. <strong>The</strong> <strong>productivity</strong> <strong>of</strong> native N + N compounds has<br />
been studied by Férnandez-Domínguez (2009), but neoclassical compounding<br />
differs from them in many ways. This gap in research might partly be due to the<br />
complex status <strong>of</strong> combining forms as a part <strong>of</strong> the English word-formation<br />
system (see, e.g., Bauer 1998): they seem to be situated somewhere between<br />
af<strong>fi</strong>xation and compounding (see section 2.2 for discussion).<br />
To make the issue even more complicated, even the OED seems to<br />
be rather inconsistent in the way it classi<strong>fi</strong>es certain elements as either combining<br />
forms or traditional af<strong>fi</strong>xes. This inconsistency seems to concern initial combining<br />
forms in particular. For example, the OED de<strong>fi</strong>nes hyper- and -ultra as pre<strong>fi</strong>xes<br />
and quasi- and pseudo- as combining forms, even though there seems to be no<br />
apparent structural difference between these two groups <strong>of</strong> elements. It has also<br />
been argued that the category <strong>of</strong> combining forms in general is superfluous and<br />
that other, less marginal types <strong>of</strong> word-formation could well be used to describe<br />
this word-formation process (Kastovsky 2009).<br />
In this paper, I will measure the <strong>morphological</strong> <strong>productivity</strong> <strong>of</strong><br />
<strong>selected</strong> English initial combining forms (ICFs) and <strong>fi</strong>nal combining forms<br />
(FCFs), three <strong>of</strong> each, and see whether there are any differences between these<br />
two rather closely related categories. <strong>The</strong> focus is on the so-called classical<br />
2
combining forms, as opposed to modern combining forms (see Prćić 2005, 2007,<br />
2008). <strong>The</strong> initial combining forms studied are hyper-, quasi-, and pseudo-, and<br />
the <strong>fi</strong>nal combining forms are -logy, -graphy, and -nomy. <strong>The</strong> study applies corpus<br />
methods, and the data comes from the written part <strong>of</strong> the British National Corpus<br />
(henceforth the BNC). My hypothesis is that the initial combining forms will<br />
prove more productive and that they will combine more easily with native bases,<br />
one <strong>of</strong> the reasons being that pre<strong>fi</strong>xes in general are <strong>of</strong>ten more lexeme-like than<br />
suf<strong>fi</strong>xes, the latter <strong>of</strong>ten being used to convey grammatical information.<br />
<strong>The</strong> main goal <strong>of</strong> this study is to clarify the status <strong>of</strong> combining<br />
forms in the English word-formation system and to provide information on their<br />
<strong>productivity</strong>. As for methodology, I will mainly rely on previous approaches that<br />
have been widely tested in several studies. I will apply several indicators to assess<br />
the <strong>morphological</strong> <strong>productivity</strong> <strong>of</strong> the <strong>selected</strong> combining forms. <strong>The</strong> <strong>fi</strong>rst one is<br />
the measure called <strong>productivity</strong> in the strict sense P (also called potential<br />
<strong>productivity</strong>), developed by Baayen and his collaborators (see Baayen and Lieber<br />
1991 in particular). In addition, two measures from lexical statistics, vocabulary<br />
growth curves and frequency spectra, are employed. An open source statistics<br />
s<strong>of</strong>tware package, R, is used to obtain information on the <strong>selected</strong> combining<br />
forms.<br />
Section two provides the reader with the theoretical background<br />
relevant for this study, de<strong>fi</strong>ning and discussing the basic concepts related to<br />
English word-formation. <strong>The</strong> category <strong>of</strong> combining forms is introduced, as well<br />
as that <strong>of</strong> neoclassical compounds. <strong>The</strong> status <strong>of</strong> combining forms as a part <strong>of</strong> the<br />
English word-formation system is <strong>of</strong> great relevance here, and it is therefore<br />
thoroughly discussed. A brief history <strong>of</strong> diachronic approaches to the study <strong>of</strong><br />
af<strong>fi</strong>xation is also introduced. <strong>The</strong> notion <strong>of</strong> <strong>morphological</strong> <strong>productivity</strong> and the<br />
theories attempting to describe it are also discussed, as well as the various<br />
constraints that affect the application <strong>of</strong> various word-formation processes. In<br />
addition, the basic theoretical background for lexical statistics and LNRE<br />
modelling are introduced. Psycholinguistic approaches to <strong>productivity</strong> are also<br />
discussed, since they have proved to capture an important aspect <strong>of</strong> the nature <strong>of</strong><br />
<strong>productivity</strong>. Furthermore, many quantitative measures <strong>of</strong> <strong>productivity</strong> are also<br />
psycholinguistically motivated.<br />
3
Section three discusses the method and the data chosen for the<br />
present study, as well as potential problems that might turn up in the research<br />
process. Important questions include the reliability <strong>of</strong> corpus evidence, as well as<br />
the principles on which the units <strong>of</strong> analysis are determined. Section four presents<br />
the results, which are then analysed and discussed in the <strong>fi</strong>fth section. Section six<br />
provides a brief summary <strong>of</strong> the most important <strong>fi</strong>ndings <strong>of</strong> the thesis and<br />
suggests some ideas for possible future studies.<br />
2. Background<br />
2.1. Word-formation<br />
2.1.1. Basic concepts <strong>of</strong> word-formation<br />
<strong>The</strong> concept <strong>of</strong> word is, in spite <strong>of</strong> its apparent simplicity, rather problematic in<br />
linguistics. <strong>The</strong> notions <strong>of</strong> lexeme and word-form have proved much more useful<br />
in the study <strong>of</strong> morphology. Haspelmath de<strong>fi</strong>nes lexemes, or dictionary words, as<br />
abstract entities that consist <strong>of</strong> several word-forms. Word-forms, or text words, on<br />
the other hand, are the concrete realizations <strong>of</strong> a lexeme that can be pronounced<br />
and used in texts (2002: 13). Plag uses the terms inflection and derivation to<br />
distinguish between word-forms and lexemes: according to him, word-forms are<br />
created with inflectional suf<strong>fi</strong>xes, while new lexemes are produced by derivational<br />
af<strong>fi</strong>xes (2003: 14; see also Kastovsky 2001: 218). Since lexemes consist <strong>of</strong> several<br />
word-forms, it is necessary to have a speci<strong>fi</strong>c citation form, under which the<br />
lexical entry is listed in dictionaries (Haspelmath 2002:<br />
14).<br />
From the point <strong>of</strong> view <strong>of</strong> <strong>morphological</strong> <strong>productivity</strong>, another<br />
central distinction in morphology, the one between possible (or potential) words<br />
and actual words, is also relevant. Plag de<strong>fi</strong>nes possible words as words whose<br />
structure follows the rules <strong>of</strong> the language. Actual words, on the other hand, are in<br />
use in the language, although it is not clear how “in use” should be de<strong>fi</strong>ned<br />
(whether it be in the vocabulary <strong>of</strong> an individual speaker, or in a dictionary) (Plag<br />
1999: 7). Plag writes that the usefulness <strong>of</strong> the possible-actual dichotomy has been<br />
4
widely criticized, but he considers it a useful notion nevertheless: according to<br />
him, studying large numbers <strong>of</strong> actual words is necessary in order to <strong>fi</strong>nd out<br />
about the mechanisms that are related to the properties <strong>of</strong> possible words (1999:<br />
8–9).<br />
<strong>The</strong> notion <strong>of</strong> mental lexicon is also relevant in the possible/actual<br />
dichotomy. Anshen and Aron<strong>of</strong>f de<strong>fi</strong>ne the speaker’s mental lexicon as “the list <strong>of</strong><br />
irregular items that the speaker/hearer carries around in his or her head” (1998:<br />
238). Words that are stored in the speaker’s mental lexicon are existing words,<br />
while potential words are words that meet all the criteria for being a genuine word<br />
in a language but that are not listed in the mental lexicon <strong>of</strong> that person (though<br />
they may exist for another speaker) (Anshen and Aron<strong>of</strong>f 1998: 238). According<br />
to Haspelmath, most morphologists agree that all simple and at least some<br />
complex words are listed in speakers’ mental lexicons, but it is dif<strong>fi</strong>cult to state<br />
which complex words are listed and which are not, because most word-formation<br />
processes cannot be described as either unequivocally productive or unproductive<br />
(2002: 41).<br />
Since this study applies corpus linguistic methods, another<br />
distinction concerning the nature <strong>of</strong> words is worth mentioning, namely that <strong>of</strong><br />
types and tokens. In a sense, types represent different word-forms and tokens<br />
different orthographic units or strings <strong>of</strong> characters in a text (see, e.g., Carstairs-<br />
McCarthy 2002: 5–6). In other words, types are distinct word-forms, and tokens<br />
all the running words in a text (Pápai 2004: 157). Thus, for example pseudoscience<br />
and pseudo-sciences are considered as separate types, even though they<br />
are strictly speaking manifestations <strong>of</strong> a single lexeme. Baayen and Lieber de<strong>fi</strong>ne<br />
tokens as all the instances <strong>of</strong> a certain type (1991: 803). Both type and token<br />
frequencies are always determined in relation to a particular corpus (Bauer 2001:<br />
47). A common notation to describe the number <strong>of</strong> tokens, i.e., corpus size, is N,<br />
while the vocabulary size, i.e., the number <strong>of</strong> types, is <strong>of</strong>ten referred to as V<br />
(Baroni 2008: 805). It must also be noted that the number <strong>of</strong> tokens in a text or in<br />
a corpus is always necessarily higher than the number <strong>of</strong> types in the same corpus.<br />
5
2.1.2. <strong>The</strong> basic notions <strong>of</strong> the English word-formation system<br />
<strong>The</strong>re are two main processes <strong>of</strong> word-formation in English, namely af<strong>fi</strong>xation by<br />
derivation and compounding. <strong>The</strong>re are other, more marginal types <strong>of</strong> wordformation<br />
in English, such as blending, clipping, and conversion, but this section<br />
will mainly concentrate on derivation and compounding, since they bear the<br />
strongest resemblance to the process <strong>of</strong> coining new words with combining forms.<br />
Af<strong>fi</strong>xes are bound morphemes that are attached to bases, 1 <strong>of</strong>ten<br />
modifying the meaning or the part <strong>of</strong> speech <strong>of</strong> the base in some way. <strong>The</strong>y are<br />
further divided into pre<strong>fi</strong>xes and suf<strong>fi</strong>xes: pre<strong>fi</strong>xes precede the base (co- in cooccur),<br />
while suf<strong>fi</strong>xes are attached at the end <strong>of</strong> the base (-less in childless).<br />
Minkova and Stockwell (2009: 101–102) state that af<strong>fi</strong>xes are relatively typical in<br />
loanwords, since almost all the Greek and Latin loanwords in English contain one<br />
or more af<strong>fi</strong>x. <strong>The</strong> third form <strong>of</strong> af<strong>fi</strong>x is called an in<strong>fi</strong>x, which is inserted into<br />
another morpheme, as in abso-bloody-lutely (Plag 2003: 11). In<strong>fi</strong>xes are generally<br />
considered as a marginal type <strong>of</strong> word-formation in English (Plag 2003: 101).<br />
With respect to semantics, the af<strong>fi</strong>xal meaning is <strong>of</strong>ten not as clear<br />
as the meaning <strong>of</strong> the root (Minkova and Stockwell 2009: 71). Biber et al. state<br />
that suf<strong>fi</strong>xes in particular <strong>of</strong>ten convey meanings that would be dif<strong>fi</strong>cult to<br />
paraphrase, a fact that might also be partially interconnected with their classchanging<br />
nature. Pre<strong>fi</strong>xes, on the other hand, <strong>of</strong>ten express meanings that could<br />
also be conveyed by using such things as adjectival premodi<strong>fi</strong>ers or other lexical<br />
words (Biber et al. 1999: 324). For example, the meaning <strong>of</strong> such suf<strong>fi</strong>xes as -ist<br />
or -ian can be explained by saying that they are used to form agentive nouns<br />
(scientist, physician). On the other hand, the meaning <strong>of</strong> pre<strong>fi</strong>xes (e.g., neo-,<br />
micro-) can <strong>of</strong>ten be explained by using adjectives (‘new’, ‘small’).<br />
<strong>The</strong>re are several ways to classify af<strong>fi</strong>xes in English, each <strong>of</strong> which<br />
has its own advantages and disadvantages. Various classi<strong>fi</strong>cations are introduced<br />
in Plag (2003: 85–86). In addition to the obvious pre<strong>fi</strong>x-suf<strong>fi</strong>x dichotomy, he<br />
mentions such criteria as the grammatical category <strong>of</strong> the base word, the semantic<br />
properties <strong>of</strong> the af<strong>fi</strong>x, and the ability <strong>of</strong> the af<strong>fi</strong>x to change the grammatical<br />
1 Root and stem are concepts that are easy to confuse, and in fact there is some disagreement as to<br />
how they interrelate. Minkova and Stockwell de<strong>fi</strong>ne root as the central element <strong>of</strong> derivational<br />
process that can either stand free (host) or be bound (seg in segment) (2009: 69). Base, on the other<br />
hand, may either contain a root or a root plus af<strong>fi</strong>x, while the term stem is usually used in<br />
inflection (2009: 71).<br />
6
category <strong>of</strong> the base word. Plag states that one <strong>of</strong> the most unproblematic<br />
classi<strong>fi</strong>cations, at least with respect to suf<strong>fi</strong>xes, is the syntactic category <strong>of</strong> the<br />
derived word. Some suf<strong>fi</strong>xes, such as -age and -ess, always result in nouns, while<br />
-ify and -ize are used to derive verbs from nouns or adjectives. Pre<strong>fi</strong>xes, on the<br />
other hand, are, because <strong>of</strong> their “more semantic” nature, <strong>of</strong>ten classi<strong>fi</strong>ed<br />
according to their meaning (see, e.g., Plag 2003: 98–99, Marchand 1969: 134,<br />
Minkova and Stockwell 2009: 102–105). As a <strong>fi</strong>nal remark, it must be noted that<br />
inflection, which allows the expression <strong>of</strong> various grammatical categories such as<br />
number (with nouns) or tense (with verbs) also makes use <strong>of</strong> af<strong>fi</strong>xes, but it is<br />
generally considered as part <strong>of</strong> syntax rather than word-formation (Plag 2003: 14).<br />
Compounding, as opposed to derivation by af<strong>fi</strong>xation, involves<br />
combining two (or possibly more) free forms into a single unit. In general,<br />
compounding is the most important source <strong>of</strong> new lexemes in the English lexicon<br />
(Minkova and Stockwell 2009: 9). As for their internal structure, compound<br />
lexemes are usually divided into endocentric and exocentric compounds. 2<br />
Endocentric compounds involve what is usually called a modi<strong>fi</strong>er-head structure,<br />
where one element, the head, is somehow modi<strong>fi</strong>ed by another element (Plag<br />
2003: 132). In English, compounds tend to be right-headed (Carstairs-McCarthy<br />
2002: 61). For example, in doll house, house is the head that is modi<strong>fi</strong>ed by doll.<br />
<strong>The</strong> head determines most <strong>of</strong> the semantic and syntactic properties <strong>of</strong> the<br />
compound as a whole, so if the head is a noun (as in doll house), the whole<br />
compound must be a noun (Plag 2003: 132).<br />
Exocentric compounds, or bahuvrihi compounds, on the other hand,<br />
exhibit a structure where the semantic head is outside the compound (Plag 2003:<br />
145). For example, pickpocket is not a kind <strong>of</strong> pocket but a person who picks<br />
pockets, and turncoat is not a kind <strong>of</strong> coat but a person betrays their original cause<br />
by shifting to the opposing side. However, as Plag points out, these compounds do<br />
have a head on the structural level, since the part <strong>of</strong> speech <strong>of</strong> the compound is<br />
again determined by the right-hand member <strong>of</strong> the compound as a whole (2003:<br />
145–146). Thus, pickpocket is a noun, not a verb. A third type <strong>of</strong> compounds<br />
2 Carstairs-McCarthy uses the terms headed and headless compounds, respectively (2002: 64).<br />
Minkova and Stockwell, on the other hand, distinguish between syntactic compounds, which are<br />
formed by regular rules <strong>of</strong> grammar and are therefore not listed in dictionaries (e.g., birthplace),<br />
and lexical compounds, the meaning <strong>of</strong> which cannot be computed from the meanings <strong>of</strong> its<br />
constituent elements (ice cream) (2009: 10).<br />
7
includes words such as singer-songwriter and poet-translator. <strong>The</strong>y are called<br />
copulative compounds, or dvandva compounds, and can be said to have two<br />
semantic heads, neither <strong>of</strong> them being more prominent than the other (Plag 2003:<br />
146).<br />
2.1.2. <strong>The</strong> diachronic perspective on the English word-formation<br />
<strong>The</strong> primary subject <strong>of</strong> this study is not the diachronic development <strong>of</strong> the English<br />
word-formation system. However, a diachronic approach may explain some <strong>of</strong> the<br />
differences between different word-formation processes in terms <strong>of</strong> <strong>morphological</strong><br />
<strong>productivity</strong> (see Carstairs-McCarthy 2002: 100). As Nevalainen has put it, the<br />
rate <strong>of</strong> language change is most easily seen in the development <strong>of</strong> lexicon, despite<br />
the fact that the core vocabulary <strong>of</strong> a language is usually relatively stable (1999:<br />
332). None <strong>of</strong> the combining forms studied here are <strong>of</strong> native origin, but belong to<br />
the foreign strata <strong>of</strong> the English lexicon.<br />
Af<strong>fi</strong>xes have originally been independent words that have had an<br />
auxiliary function, <strong>of</strong>ten derivational or inflectional (Warren 1990: 123–124).<br />
Eventually, they have gone through a grammaticalization process, which has<br />
resulted in semantic and phonetic erosion <strong>of</strong> varying degrees. Minkova and<br />
Stockwell call this phenomenon semantic bleaching (2009: 102). For example, the<br />
original meaning <strong>of</strong> the suf<strong>fi</strong>x -hood (‘condition’, ‘state <strong>of</strong> affairs’) is almost<br />
completely bleached out and is almost impossible to specify in contemporary<br />
English. Other similar examples in modern English include the native suf<strong>fi</strong>xesdom<br />
and -ship, which are derived from nouns (Minkova and Stockwell 2009: 72).<br />
On the other hand, sometimes the meaning <strong>of</strong> a word develops in the opposite<br />
direction and a bound morpheme may acquire a new meaning as a free root, like<br />
the noun ex (‘a former spouse’) (Minkova and Stockwell 2009: 72).<br />
In general, the assumption is that new non-native af<strong>fi</strong>xes are born<br />
when a suf<strong>fi</strong>cient number <strong>of</strong> complex lexemes are borrowed into a language and<br />
they thus start to be interpreted as <strong>morphological</strong>ly transparent (Nevalainen 1999:<br />
377, Marchand 1969: 129). Transparency, which means the fact that a lexeme is<br />
8
formally and semantically analysable into its constituent parts, seems to be at least<br />
indirectly linked to <strong>productivity</strong> (Dalton-Puffer 1994: 252–253, Pounder 2000:<br />
134). <strong>The</strong> relationship between transparency and <strong>productivity</strong> is discussed in more<br />
detail in section 2.3.1.<br />
In the course <strong>of</strong> history, the English vocabulary has been shaped by<br />
several waves <strong>of</strong> loanwords from various languages, such as French, Latin, Greek,<br />
and the so-called Neo-Latin, which was a mixture <strong>of</strong> Latin and Greek vocabulary<br />
and which served as the lingua franca during the Middle Ages and the<br />
Renaissance (Kastovsky 2009: 1). Besides the vocabulary, this considerable<br />
borrowing has also affected the patterns <strong>of</strong> derivational morphology available to<br />
speakers (Cowie and Dalton-Puffer 2002: 410).<br />
In terms <strong>of</strong> such <strong>morphological</strong> strata, the lexicon <strong>of</strong> Old English<br />
was very homogeneous, with only about three percent <strong>of</strong> the lexical items being <strong>of</strong><br />
non-native origin (Kastovsky 2006: 167). <strong>The</strong> Norman Conquest, however,<br />
brought about enormous changes in the English word-formation system, and<br />
many Germanic word-formation processes became obsolete (Marchand 1969:<br />
130). <strong>The</strong>se non-native derivational patterns became productive gradually<br />
(Kastovsky 2006: 167–168). First, individual lexical items were borrowed, and<br />
after a certain point “a formal-semantic relationship” could be established so that<br />
the pattern could be extended to new formations by analogy, until the wordformation<br />
process <strong>fi</strong>nally became productive.<br />
<strong>The</strong> number <strong>of</strong> native pre<strong>fi</strong>xes in contemporary English is very small<br />
(Marchand 1969: 129). <strong>The</strong>y include a-, be-, fore-, mid-, mis-, and un-, and they<br />
have their origin in independent words. During the Middle English period, a great<br />
number <strong>of</strong> af<strong>fi</strong>xes, such as dis-, en-, inter-, non-, -age, -ation, -ive and -ous, was<br />
borrowed from French and Latin, and by the Early Modern English period they<br />
had become very frequent (Kastovsky 2006: 168). This abundance <strong>of</strong> borrowing<br />
affected the whole word-formation system by causing competition, not only<br />
between native and non-native af<strong>fi</strong>xes, but also between the non-native elements<br />
themselves (e.g. the various negative pre<strong>fi</strong>xes a-, dis-, in- and non-) (Kastovsky<br />
2006: 169). As for suf<strong>fi</strong>xes, the effect <strong>of</strong> the invasion <strong>of</strong> French, Latin and Greek<br />
loanwords has not been as radical (Marchand 1969: 227).<br />
During the Modern English period, the extensive borrowing has<br />
continued, this time because <strong>of</strong> the development <strong>of</strong> scienti<strong>fi</strong>c terminology,<br />
9
introducing elements such as bi-, di-, hyper-, hypo-, meta-, micro- and multi- – the<br />
number <strong>of</strong> pre<strong>fi</strong>xes outnumbering that <strong>of</strong> suf<strong>fi</strong>xes (Kastovsky 2006: 170). In<br />
addition, any learned af<strong>fi</strong>xes, or combining forms, 3 have acquired new meanings<br />
(Hughes 2004: 347). For example, extra- used to mean ‘out <strong>of</strong>’, as in<br />
extraordinary (‘outside the scope <strong>of</strong> ordinary’) but is nowadays used to denote a<br />
meaning ‘more than the usual’, as in extra strong. Similarly, super- ‘above’ and<br />
ultra- ‘beyond’ are increasingly used to stress<br />
size.<br />
Kastovsky summarizes that the overall structure <strong>of</strong> English<br />
vocabulary has changed and the number <strong>of</strong> derivational patterns available has<br />
increased enormously from Old English to the present day, “partly rivalling each<br />
other, partly restricting each other according to etymological domains” (2006:<br />
170). In a similar vein, Hughes writes that different word-formation patterns are<br />
used with greater flexibility in contemporary English than they were before:<br />
recent coinages that exploit the suf<strong>fi</strong>x -able include, for example, such words as<br />
doable, puttable and liveable (2000: 346).<br />
2.2. Combining forms<br />
2.2.1 De<strong>fi</strong>ning combining forms<br />
It was stated above that af<strong>fi</strong>xes are always bound morphemes that attach to a root<br />
or a base. What should then be thought about such words as neuro-logy or biology?<br />
At <strong>fi</strong>rst sight it seems that lexemes like these are formed with a pre<strong>fi</strong>x and a<br />
suf<strong>fi</strong>x but no root, which <strong>of</strong> course runs counter to the basic assumption that<br />
words should always contain a root (see Bauer 1983: 213–214).<br />
Carstairs-McCarthy <strong>of</strong>fers a partial solution by stating that af<strong>fi</strong>xes<br />
are always bound but roots are not always free (2002: 20). Thus the <strong>fi</strong>rst element<br />
in words like audi-ence, magn-ify, or applic-ant is actually a bound root, since<br />
they cannot occur alone in English (Castairs-McCarthy 2002: 19). <strong>The</strong> common<br />
denominator with all these words is that they are all loanwords, borrowed either<br />
3 <strong>The</strong> distinction between af<strong>fi</strong>xes and combining forms will be discussed in more detail in section<br />
2.2.<br />
10
from Latin or Latin via French. 4 <strong>The</strong> same “boundedness” applies to such<br />
formations as neurology and biology as well, but in this case not only the <strong>fi</strong>rst<br />
element but also the <strong>fi</strong>nal constituent is bound. This is what Castairs-McCarthy<br />
means when he points out that words can in fact consist <strong>of</strong> more than one bound<br />
root (2002: 21). In this sense, the constituents in words like neurology and biology<br />
could also be called bound roots, but it has become a common practice by<br />
linguists and dictionary-makers to call them neoclassical elements, or, more<br />
commonly, combining forms (Plag 2003: 74; Carstairs-McCarthy 2002: 21).<br />
<strong>The</strong> term combining form was <strong>fi</strong>rst used in the New English<br />
Dictionary (1884–1928), which was the predecessor <strong>of</strong> the Oxford English<br />
Dictionary (Kastovsky 2009: 2; see McCauley 2006 for the use <strong>of</strong> the term<br />
combining form in different editions <strong>of</strong> NED and OED). <strong>The</strong> current version <strong>of</strong><br />
the OED Online has a total <strong>of</strong> 2244 entries that are categorized as combining<br />
forms. However, many entries under the label <strong>of</strong> combining forms have been <strong>fi</strong>rst<br />
published as early as in 1899, but have not yet been fully updated. As for the<br />
de<strong>fi</strong>nition <strong>of</strong> the term itself, the OED Online cites Bloch and Trager (1942: 66):<br />
In Latin and other languages, many words have a special<br />
combining form which appears only in compounds (or<br />
only in compounds and derivatives)... <strong>The</strong> foreign-learned<br />
part <strong>of</strong> the English vocabulary also shows a number <strong>of</strong><br />
special combining forms; cf. electro-, combining form <strong>of</strong><br />
electric, in such compounds as electromagnet.<br />
For the third edition <strong>of</strong> the OED, the category <strong>of</strong><br />
combining forms is de<strong>fi</strong>ned in the following way:<br />
[A]n element used, either initially or <strong>fi</strong>nally, in<br />
combination with another element to form a word. For the<br />
purposes <strong>of</strong> OED3, a combining form differs from a pre<strong>fi</strong>x<br />
or suf<strong>fi</strong>x by being generally noun-like or adjective-like<br />
4 It is not always easy to differentiate between loanwords <strong>of</strong> Latin and French origin, partly<br />
because these two languages are etymologically related and because it is not always clear at which<br />
point in history a certain word has been borrowed into English. When borrowing from<br />
Latin/French was at its most active phase, England was in fact a trilingual community, which<br />
makes it even trickier to trace the etymology <strong>of</strong> some words (see McConchie 2006). Nevalainen in<br />
fact uses the term latinate to describe words that are <strong>of</strong> either Latin or French origin (1999: 371).<br />
11
and having a relatively full lexical meaning. (McCauley<br />
2006: 96)<br />
Marchand writes that the practice <strong>of</strong> coining new lexemes with<br />
learned pre<strong>fi</strong>xes and combining forms dates back to the 16 th century (1969: 131).<br />
This was mainly due to the Renaissance and the revival <strong>of</strong> sciences in Western<br />
Europe, which produced an increasing need for scienti<strong>fi</strong>c vocabulary (see also<br />
Carstairs-McCarthy 2002: 66). However, the practice <strong>of</strong> coining new words with<br />
combining forms did not become common until the 18 th and 19 th centuries, along<br />
with the industrial revolution and the development <strong>of</strong> modern science that it<br />
brought about (Marchand 1969: 131; Carstairs-McCarthy 2002: 66–67).<br />
In general, the proportion <strong>of</strong> specialist terms in the lexical intake<br />
grew steadily throughout the eighteenth century, which contributed to the fact that<br />
neoclassical formations started to be associated with technical registers<br />
(Nevalainen 1999: 365). <strong>The</strong> development <strong>of</strong> learned vocabulary usually goes<br />
hand in hand with the development <strong>of</strong> the corresponding domains <strong>of</strong> science and<br />
scholarship (Fradin 2000: 23). Quirk et al. point out that in Present-Day English,<br />
in addition to scienti<strong>fi</strong>c registers, combining forms are also common in<br />
commercial brand names aimed at international markets (1985: 1535). Many<br />
neoclassical compounds are also “in international currency”, which means that<br />
they have been adopted in some form in several languages (Quirk et al. 1985:<br />
1575).<br />
It must be noted that there is some variation in terminology. So far<br />
we have only talked about what has been usually called classical combining forms<br />
(see, e.g., Fradin 2000: 11; Prćić 2005: 317). However, the term combining form<br />
has also been extended to cover elements such as -gate in Watergate or -ware in<br />
shareware (Lehrer 1990: 14, Fradin 2000: 11). Lehrer calls them “productive<br />
splinters”, claiming that they have more in common with combining forms than<br />
with af<strong>fi</strong>xes or roots (1990: 14). Prćić distinguishes between classical combining<br />
forms, which have been adopted from Greek or Latin, and modern combining<br />
forms, which are based on modern English words but which have been modi<strong>fi</strong>ed<br />
to look like combining forms by, for example, the addition <strong>of</strong> a thematic vowel or<br />
by clipping (2005: 317; 2007: 383). Examples include jazzo-phile) (from jazz), or<br />
Thatcher-nomics (from economics). Another type <strong>of</strong> formation that Prćić de<strong>fi</strong>nes<br />
12
as combining forms includes various other bound forms that are independent<br />
words in their source languages (e.g., scandal-monger and dulls-ville) (2007:<br />
384). Prćić calls combinations like these, where one element is modern and the<br />
other neoclassical (either with modern ICF + neoclassical FCF or neoclassical ICF<br />
+ modern base) “semi-neoclassical compositions” (2005: 321).<br />
This study will concentrate on what has been called classical<br />
combining forms, which are said to represent the “core <strong>of</strong> established combining<br />
forms” and are usually listed under the label <strong>of</strong> combining forms in dictionaries as<br />
well (Fradin 2000: 12). Warren de<strong>fi</strong>nes them as allomorphic variants <strong>of</strong> Greek or<br />
Latin model words (1990: 115). <strong>The</strong>y can either be derived from lexemes proper<br />
(-logy) or Greek or Latin prepositions (hyper-, quasi-). Due to the limited scope <strong>of</strong><br />
this study, I have therefore excluded formations that are clippings from preexisting<br />
words (e.g., euro-) and elements that are not strictly speaking new<br />
morphemes at all (e.g., -gate; see Warren 1990: 115). <strong>The</strong> etymology <strong>of</strong> the<br />
<strong>selected</strong> combining forms is discussed in more detail in section 3.1.<br />
Bauer distinguishes between initial combining forms (ICF) and <strong>fi</strong>nal<br />
combining forms (FCF), depending on whether they occur in initial or <strong>fi</strong>nal<br />
position (2003: 214). This dichotomy has become widely adopted in studies<br />
concentrating on combining forms (see, e.g., Kastovsky 2009). Quirk et al. write<br />
that “the <strong>fi</strong>rst element [<strong>of</strong> a compound] may be in its special ‘combining form’ (as<br />
in the noun trouserleg or the adjective socioeconomic)”, and thus exclude <strong>fi</strong>nal<br />
combining forms from this de<strong>fi</strong>nition (1985: 1567). In a similar vein, some<br />
editions <strong>of</strong> the OED reserve the term combining form for the <strong>fi</strong>rst element <strong>of</strong> a<br />
compound and use the term terminal element for the <strong>fi</strong>nal element (Kastovsky<br />
2009: 4). However, this distinction has generally been replaced by the one<br />
between ICFs and FCFs, which will be adopted in this study as well.<br />
2.2.2. Neoclassical compounds<br />
Formations that make use <strong>of</strong> combining forms are called neoclassical compounds<br />
(Plag 2003: 74). Bauer mentions several characteristics that neoclassical<br />
compounds have in common with native compounds in English (1998: 405). For<br />
example, the meaning relationship between the constituent parts varies and cannot<br />
13
always be “deduced in isolation from its context”. Secondly, they follow the<br />
condition <strong>of</strong> hyponymy in the sense that they are hyponyms <strong>of</strong> their head element<br />
(which is the rightmost element <strong>of</strong> the compound). <strong>The</strong>se same word-formation<br />
strategies were functional in the donor languages (Greek and occasionally Latin)<br />
as well, the only difference thus being the fact that neoclassical compounds are<br />
modern formations that are not attested in Greek and Latin as such (Bauer 1998:<br />
405).<br />
What makes neoclassical compounds atypical in English is that the<br />
English language seems to prefer compounds that consist <strong>of</strong> free roots, such as<br />
bookcase or motorbike (Carstairs-McCarthy 2002: 21). Combining forms, on the<br />
other hand, have “little or no currency” as separate words, and they are not<br />
normally the stressed part <strong>of</strong> a complex word (Quirk et al. 1985: 1520). Since they<br />
consist <strong>of</strong> bound elements but otherwise behave like compounds, they have<br />
sometimes been described as situated between compounding and af<strong>fi</strong>xation in the<br />
English word-formation system (Quirk et al. 1985: 1520).<br />
Many neoclassical compounds contain a central linking vowel,<br />
usually -o-, 5 which cannot unequivocally be assigned to either <strong>of</strong> the constituent<br />
combining forms (Carstairs-McCarthy 2002: 66). If the vowel is considered as<br />
part <strong>of</strong> the <strong>fi</strong>rst constituent, the <strong>fi</strong>nal combining form in such words as<br />
anthropology and cosmology will be -logy. However, in such words as toyology<br />
(two occurrences in the BNC) or grandadology (one occurrence), the initial<br />
element is <strong>of</strong> native origin, and thus the <strong>fi</strong>nal element might be considered to be -<br />
ology, which could then be considered as a <strong>morphological</strong>ly conditioned<br />
allomorph <strong>of</strong> -logy, in which case the central -o- will be part <strong>of</strong> the <strong>fi</strong>nal<br />
constituent (Kastovsky 2009: 6).<br />
<strong>The</strong> primary stress <strong>of</strong> neoclassical compounds is usually on this<br />
central -o-, a feature that distinguishes them from native compounds that usually<br />
have the stress on the <strong>fi</strong>rst element (Carstairs-McCarthy 2002: 66). In the majority<br />
<strong>of</strong> cases, the -o- does not show up if the initial combining form ends with a vowel<br />
or the <strong>fi</strong>nal combining form begins with a vowel (e.g., tele-scope vs. laryngoscope)<br />
(Plag 2003: 158). However, in some cases, such as in bio- and geo-, the -o-<br />
5 Since the linking vowel -a- has been gaining ground especially in modern formations, Prćić calls<br />
it a “modern linking vowel” (2007: 384). It must also be noted that the orthographically, other<br />
forms are accepted as well, provided the appropriate phonetic criteria are met, as in bureaucracy.<br />
14
seems to be part <strong>of</strong> the inherent structure <strong>of</strong> the combining form, hence bio-energy<br />
and geoarchaeological (Plag 2003: 158). According to Kastovsky, the vowel has<br />
originally been either a stem-formative or an inflectional ending (and thus<br />
associated with the <strong>fi</strong>rst element), but they have both become opaque in English<br />
(2009: 6, 10). He suggests that the most useful interpretation <strong>of</strong> -o- would be as a<br />
simple linking vowel.<br />
Quirk et al. note that compounding with initial combining forms<br />
may also be recursive, i.e., involve more than two elements, such as in<br />
neurolymphomatosis or nephrolithotomy (1985: 1576). Final combining forms, on<br />
the other hand, cannot be used recursively (Prćić 2007: 384). In terms <strong>of</strong><br />
semantics, the meaning <strong>of</strong> neoclassical compounds is usually predictable from the<br />
meaning <strong>of</strong> its constituent combining forms (anthrop(o)- ‘human’ and , -(o)logy<br />
‘science or study’ equal ‘science or study <strong>of</strong> human beings’), which is important<br />
especially in coining new technical terms (Carstairs-McCarthy 2002: 66).<br />
<strong>The</strong>re are three ways in which a neoclassical compound may enter a<br />
language (Cottez 1985: XVII, cited in Fradin 2000: 24). <strong>The</strong> lexeme may already<br />
exist in the donor language, in which case it is borrowed with the same meaning<br />
and corresponding form (phlebotomy vs. Greek φλεβοτομία). <strong>The</strong> second way is<br />
adaptation: the lexeme exists in the donor language, but with a different meaning<br />
(physiology vs. Greek φυσιολογία ‘natural sciences’). <strong>The</strong> third way is creating<br />
new lexemes by combining elements that exist in the donor language as separate<br />
units, not as a whole (chronology from χρονο and λογία).<br />
2.2.3. <strong>The</strong> status <strong>of</strong> combining forms<br />
Bauer argues that there is plenty <strong>of</strong> evidence for de<strong>fi</strong>ning combining forms as a<br />
category distinct from af<strong>fi</strong>xes: combining forms can combine with each other (an<br />
ICF + FCF) but not with affxes, and they usually look more like lexical items than<br />
af<strong>fi</strong>xes (1983: 214–215). <strong>The</strong> latter feature also applies to their phonological<br />
structure, since they carry their own stress, contain full vowels, and are generally<br />
disyllabic (Bauer 2003a: 35). In addition, af<strong>fi</strong>xes constitute a closed set <strong>of</strong><br />
morphemes, while combining forms form a relatively open set that grows when<br />
15
new lexemes are created with the help <strong>of</strong> resources from Greek and Latin – or<br />
English (Prćić 2005: 316).<br />
Yet another feature that has been suggested as distinguishing<br />
combining forms and af<strong>fi</strong>xes is that the same af<strong>fi</strong>x can never function as both<br />
pre<strong>fi</strong>x and af<strong>fi</strong>x, while some combining forms can be seen both as ICF and FCF<br />
(e.g., patho- in pathology -pathy in cardiopathy, both from Greek πάθος<br />
‘emotion’) (Harastani et al. 2012: 74). On the other hand, combining forms also<br />
have plenty <strong>of</strong> shared features with traditional af<strong>fi</strong>xes: for example, both can be<br />
separated from the remaining element <strong>of</strong> the lexeme without their meaning or<br />
form being changed, and both produce “output formations <strong>of</strong> a binary structure”<br />
that are <strong>morphological</strong>ly and semantically analysable into their parts (Prćić 2005:<br />
315–316).<br />
<strong>The</strong> line between combining forms and other non-native bound roots<br />
is not always clear either: some combining forms can be combined with<br />
traditional af<strong>fi</strong>xes as well (Carstairs-McCarthy 2002: 66). For example, socio- and<br />
-logy can be used with the suf<strong>fi</strong>xes -al and -ic(al) to form such words as (psycho-<br />
)social and logic(al). Prćić calls this phenomenon suf<strong>fi</strong>xal expansion, stating that<br />
in such cases, the combining form becomes composite in nature. <strong>The</strong> added suf<strong>fi</strong>x<br />
becomes the head <strong>of</strong> the combining form, and the word as a whole can be seen to<br />
display a dual headedness where the head at the CF level (e.g., -ic) can be termed<br />
the micro-head and the head at the word level (logic) a macro-head (2007: 386–<br />
387). In this study, such suf<strong>fi</strong>xal expansions will not be taken into account in<br />
analysing the <strong>selected</strong> <strong>fi</strong>nal combining forms, for reasons explained in section 3.3.<br />
To summarize, the status <strong>of</strong> combining forms in the English wordformation<br />
system is not clear. <strong>The</strong>y have features that distinguish them from<br />
compounding on the native basis, but they are also distinct from af<strong>fi</strong>xes. <strong>The</strong><br />
problem is that neoclassical combining forms have not been studied very<br />
systematically (see Kastovsky 2009: 1). Marchand criticizes the OED for being<br />
inconsistent with the term and for not properly discussing their exact role in<br />
English word-formation (1969: 132, 218). What he claims is certainly true, and<br />
the OED’s, de<strong>fi</strong>nitions for the category itself are vague at best.<br />
However, Marchand himself is rather inconsistent in de<strong>fi</strong>ning<br />
combining forms and does not provide any solution as to their status. For<br />
example, he writes that elements that are attached to full English words should be<br />
16
termed pre<strong>fi</strong>xes, and others “are <strong>of</strong> a purely dictionary interest in any case” (1969:<br />
132). This, as Kastovky has noted, is “a very questionable argument”, since they<br />
are extremely frequent, especially in scienti<strong>fi</strong>c registers (2009: 4).<br />
Bauer argues that neoclassical compounding is in fact a rather<br />
arbitrary category (they can be a part <strong>of</strong> clippings and blends, like in telethon, a<br />
blend <strong>of</strong> television and marathon) and that it should be de<strong>fi</strong>ned in terms <strong>of</strong> more<br />
or less prototypical members <strong>of</strong> the category (1998: 407–409, 419; see also<br />
Kastovsky 2009: 8). He suggests that problems with classifying neoclassical<br />
compounds and combining forms might be due to the falsely unambiguous label<br />
“neoclassical compound” itself (1998: 414). Kastovsky presents a similar view,<br />
but goes even further by claiming that the term might as well be abandoned, since<br />
“modern word-formation theory can well do without it” (2009: 2). He explains<br />
this by the fact that even though modern English word-formation is dominated by<br />
word-based morphology, 6 English has recently adopted some stem-based<br />
morphology as well in the form <strong>of</strong> borrowed non-native elements, such as the<br />
soci-al and logi-al mentioned above (2009: 9).<br />
His solution thus involves simply introducing the concept <strong>of</strong> stem as<br />
a possible lexeme representation (2009: 11), thus elaborating on the ideas<br />
presented by Carstairs-McCarthy (2002: 21). <strong>The</strong> stems can either combine with<br />
af<strong>fi</strong>xes (electr-ify) or with another stem (socio-logy), where the latter process<br />
produces what has been called neoclassical compounds. As for the modern<br />
combining forms, Kastovsky writes that such established word-formation<br />
processes as blending (chunnel, from channel + tunnel) and clipping (euro-, as in<br />
Eurocentric, eurocrat) can be used to account for them (2009: 11–12).<br />
Kastovsky’s proposition is certainly appealing, and would make the<br />
make-up <strong>of</strong> the English word-formation system much more simple and consistent.<br />
It is, however, the ability <strong>of</strong> combining forms to combine with af<strong>fi</strong>xes that causes<br />
some trouble. Bauer presents such ungrammatical examples as *electroness and<br />
*electronization to justify the claim that the combinations are in general not<br />
possible. It is true that such combinations electr-ic or soci-al are perfectly<br />
grammatical, which would indicate the contrary. It must be also kept in mind that<br />
6 In word-based morphology, the base form is directly accessed and thus not an abstraction, while<br />
in stem-based morphology, the base does not occur without a derivational (or inflectional)<br />
morpheme, such as in scient-ist (Kastovsky 2006: 157).<br />
17
not all af<strong>fi</strong>xes can attach to all combining forms, just as it is not possible for any<br />
af<strong>fi</strong>x to attach to any base. However, there are also some (usually initial)<br />
combining forms that do not seem to be able to combine with any af<strong>fi</strong>xes, such as<br />
pseudo- and hyper-. This might indicate that there is a cline between more af<strong>fi</strong>xlike<br />
and more lexeme-like combining forms.<br />
In addition, Kastovsky himself seems to be somewhat inconsistent<br />
by terming such elements as auto-, hypo- and mono- as pre<strong>fi</strong>xes instead <strong>of</strong><br />
combining forms. But what is the exact quality that would make them pre<strong>fi</strong>xes,<br />
not combining forms? <strong>The</strong> OED de<strong>fi</strong>nes auto- and mono- as combining forms,<br />
and hypo- as a pre<strong>fi</strong>x (the former being developed from lexical words, the latter<br />
from a preposition).<br />
If Kastovsky’s view is accepted, it will raise a number <strong>of</strong> questions.<br />
What is the bene<strong>fi</strong>t <strong>of</strong> studying the <strong>productivity</strong> <strong>of</strong> a category that does not exist in<br />
its own right? Still, the category <strong>of</strong> combining forms (at least in the classical<br />
sense) might be useful from the point <strong>of</strong> view <strong>of</strong> the study <strong>of</strong> scienti<strong>fi</strong>c<br />
terminology. Furthermore, the <strong>productivity</strong> <strong>of</strong> these <strong>selected</strong> elements has not<br />
been studied yet, and it may still reveal something new about them and help<br />
situate them in the big picture <strong>of</strong> the English word-formation system. In addition,<br />
it is the differences between initial and <strong>fi</strong>nal combining forms that we are<br />
interested in, and in this kind <strong>of</strong> study a term like combining form might not be<br />
irrelevant. All in all, it can be said that none <strong>of</strong> the suggested solutions concerning<br />
the status <strong>of</strong> combining forms presented in literature has proved completely<br />
satisfactory so far. Thus, the term combining form will be used in this study,<br />
partially for lack <strong>of</strong> a better term.<br />
2.3. Morphological <strong>productivity</strong><br />
2.3.1. De<strong>fi</strong>ning <strong>morphological</strong> <strong>productivity</strong><br />
Morphological <strong>productivity</strong> is an important concept in the study <strong>of</strong> wordformation<br />
and <strong>morphological</strong> theory in general (Cowie and Dalton-Puffer 2002:<br />
411). Productivity is <strong>of</strong>ten de<strong>fi</strong>ned as the speakers’ ability to create new lexemes<br />
by exploiting the resources <strong>of</strong> a language. Speakers are able to create an almost<br />
unlimited number <strong>of</strong> new words, and the lexicon <strong>of</strong> a language is thus never <strong>fi</strong>xed<br />
18
(Haspelmath 2002: 39). <strong>The</strong>refore, word-formation processes can be called either<br />
productive or unproductive, depending on the extent to which they can be applied<br />
to form new <strong>morphological</strong>ly complex words (Plag 2006: 537). 7<br />
To illustrate the difference between productive and unproductive<br />
word-formation processes, Plag mentions the verb-forming suf<strong>fi</strong>x -ize<br />
(grammaticalize) and the pre<strong>fi</strong>x en- (enlist), which are both used to form verbs in<br />
English (2006: 537). <strong>The</strong> former can be considered fairly productive, since it is<br />
frequently used to form new lexemes that have not yet made their way in the<br />
established lexicon. <strong>The</strong> latter, on the other hand, cannot be so easily used in new<br />
formations in contemporary English, and it is thus felt to be unproductive. 8<br />
Haspelmath writes that unproductive rules are a unique property <strong>of</strong><br />
morphology, since there is no direct equivalent to them in the area <strong>of</strong> syntax<br />
(2002: 40). Carstairs-McCarthy explains this phenomenon by mentioning the<br />
nonexistence <strong>of</strong> clearly ill-formed words (*ion-trans-al-at-form), but also <strong>of</strong> some<br />
words that are in theory well-formed (*arrivation, *ridiculosity), as well as the<br />
unpredictability and non-compositionality <strong>of</strong> the meaning <strong>of</strong> some existing words<br />
(recital vs. recitation) (1992: 32). Bauer states that it is the <strong>productivity</strong> <strong>of</strong> wordformation<br />
processes that is at least partially responsible for the huge vocabulary <strong>of</strong><br />
English, a fact that can be veri<strong>fi</strong>ed by consulting various dictionaries <strong>of</strong><br />
neologisms (1983: 63).<br />
Another way to illustrate the difference between productive and<br />
unproductive categories comes from Baayen, who de<strong>fi</strong>nes productive categories<br />
as ones with a growing membership, while unproductive categories have a <strong>fi</strong>xed<br />
or declining membership (2009: 900). Whether <strong>productivity</strong> should be understood<br />
as an essentially synchronic or diachronic phenomenon (or both), is controversial,<br />
however. Bauer claims that it is possible to talk about <strong>productivity</strong> in synchronic<br />
terms, or about changes in <strong>productivity</strong> in diachronic terms, but not <strong>of</strong><br />
<strong>productivity</strong> as such as a diachronic phenomenon (1988: 61). Cowie and Dalton-<br />
7 Terms such as word-formation process, word-formation rule (<strong>of</strong>ten abbreviated as WFR),<br />
<strong>morphological</strong> rule, and <strong>morphological</strong> category are <strong>of</strong>ten used more or less synonymously in the<br />
literature. <strong>The</strong>y are usually used to refer to the way morphemes can be combined to form lexemes<br />
(Plag 2003: 30, 179). <strong>The</strong> rule must contain information at least about the phonology and<br />
semantics <strong>of</strong> the morpheme as well as about possible base morphemes (Plag 2003: 31). This is<br />
what distinguishes “rules” or “processes” from the rather mechanical concept <strong>of</strong> morpheme. This<br />
study will mainly use the term word-formation process.<br />
8 One <strong>of</strong> the few exceptions is encrypt, which was <strong>fi</strong>rst attested in the 1950’s, according to the<br />
OED.<br />
19
Puffer write that if <strong>productivity</strong> is to be taken as “a design feature <strong>of</strong> language in<br />
general”, it should be understood as a synchronic notion (2002: 417).<br />
On the other hand, it has been suggested that using type frequency,<br />
for example, as an indicator <strong>of</strong> <strong>productivity</strong> (i.e., counting all the types formed<br />
between two points in time) is inherently diachronic, even though the aim was to<br />
study <strong>productivity</strong> from a synchronic point <strong>of</strong> view (see Cowie and Dalton-Puffer<br />
2002: 418 for discussion). Be that as it may, changes in the <strong>productivity</strong> <strong>of</strong> certain<br />
word-formation patterns have been frequently studied as well (see, e.g., Dalton-<br />
Puffer 1994, 1996; Cowie 1998; Cowie and Dalton-Puffer 2002; and Säily and<br />
Suomela 2009).<br />
Plag asks whether <strong>productivity</strong> should be seen as a theoretical<br />
primitive (i.e., an inherent property <strong>of</strong> word-formation rules) or whether it results<br />
from other properties <strong>of</strong> these rules (2006: 537). According to Dressler and<br />
Ladányi, <strong>productivity</strong> is “a primitive and prototypical property” <strong>of</strong> wordformation<br />
rules, comparable to inflectional, syntactic, or phonological rules (2000:<br />
119). On the other hand, Férnandez-Domínguez (2009: 53) claims that since<br />
linguists <strong>of</strong>ten list a number <strong>of</strong> features associated with <strong>productivity</strong> (e.g., the<br />
various constraints on it), the phenomenon might be considered as being <strong>of</strong><br />
composite nature and not an indivisible notion.<br />
<strong>The</strong> distinction between inflection and derivation is also relevant in<br />
de<strong>fi</strong>ning <strong>morphological</strong> <strong>productivity</strong>. Although <strong>productivity</strong> is <strong>of</strong>ten considered<br />
more relevant for the study <strong>of</strong> word-formation rather than that <strong>of</strong> inflectional<br />
processes, Plag states that, at least to some extent, <strong>productivity</strong> also has to do with<br />
inflection (2006: 538). For example, the regular past tense af<strong>fi</strong>x -ed might not be<br />
considered fully productive, since some verbs form past tense forms with ablaut<br />
or take no overt suf<strong>fi</strong>x, although the formation <strong>of</strong> the past tense is fully productive<br />
as a category. According to Bauer, concatenative processes such as conversion<br />
and back-formation should also be understood as <strong>morphological</strong> processes and<br />
thus relevant for <strong>productivity</strong> as well (2005: 316). <strong>The</strong> <strong>productivity</strong> <strong>of</strong> compounds<br />
has also been researched, though to a lesser extent (see Férnandez-Domínguez<br />
2009). Bauer goes as far as to suggest that <strong>productivity</strong> even concerns syntax:<br />
some syntactic structures are used more frequently than others to form new<br />
sentences, and thus the notion <strong>of</strong> syntactic markedness might be understood as<br />
some form <strong>of</strong> <strong>productivity</strong> or as relating to it (1995: 21).<br />
20
Morphological <strong>productivity</strong> and linguistic creativity are two<br />
concepts that are easy to confuse. <strong>The</strong>re is in fact some confusion and overlapping<br />
between the two terms among linguists themselves (see, e.g., Bauer 2001: 1, 62ff).<br />
According to Bauer, <strong>productivity</strong> is an inherent feature <strong>of</strong> human language, which<br />
allows speakers to exploit the rules <strong>of</strong> word-formation to create new lexemes,<br />
while creativity allows for a native speaker “to extend the language system in a<br />
motivated, but unpredictable (non-rule-governed) way” (1983: 63; see also Booij<br />
2005: 67 and Dressler and Ladányi 2000: 105).<br />
Renouf writes that creativity is <strong>of</strong>ten employed to achieve a stylistic<br />
effect, and is typically seen as a departure from the rules <strong>of</strong> grammar, which can<br />
manifest itself in such instances as punning or wilful error (2007: 70). Evert and<br />
Lüdeling state that words that result from productive processes are <strong>of</strong>ten not<br />
recognized as new by speakers, while creative formations are perceived as new<br />
(2001: 475). Bauer also uses the term “semi-<strong>productivity</strong>” to describe linguistic<br />
creativity, claiming that it differs from <strong>productivity</strong> by requiring some sort <strong>of</strong><br />
encyclopaedic, extra-linguistic knowledge (1995: 26, 28). Plag writes that<br />
<strong>productivity</strong> is <strong>of</strong>ten considered unintentional and creativity intentional, but he<br />
also claims that the intentional/unintentional dichotomy is not very useful, since<br />
speakers have varying levels <strong>of</strong> awareness <strong>of</strong> the structure <strong>of</strong> the lexicon. In<br />
addition, even productive rules may be used intentionally (1999: 13–14).<br />
It must be noted, however, that <strong>productivity</strong> should not be confused<br />
with regularity either, since all productive patterns are rule-governed and thus<br />
regular, but so are many unproductive rules as well (Dressler and Ladányi 2000:<br />
105). For example, the suf<strong>fi</strong>x -ess is perfectly regular, but cannot be used to coin<br />
neologisms (Haspelmath 2002: 111). <strong>The</strong> same holds true for transparency: a<br />
lexeme can be perfectly transparent but not necessarily productive (Pounder 2000:<br />
134–135; Bauer 2001: 54). 9 On the other hand, productive processes are usually<br />
formally and semantically transparent, unless the process becomes lexicalized<br />
(Bauer 2001: 135). Dalton-Puffer points out that the term transparency is <strong>of</strong>ten<br />
9 Transparency and regularity are closely related concepts, and sometimes used synonymously.<br />
Pounder (2000: 135) suggests that transparent emphasizes the “analysis side <strong>of</strong> morphology”,<br />
while regular has more to do with the fact that the formation is not isolated in the <strong>morphological</strong><br />
system <strong>of</strong> a language from the speaker’s point <strong>of</strong> view. Another related concept is analysability,<br />
which is used by Dalton-Puffer 1994, who uses it more or less synonymously with transparency.<br />
21
used in a vague sense in the literature, since it is usually not speci<strong>fi</strong>ed whether the<br />
focus is on the linguist’s or the speaker’s point <strong>of</strong> view (1994: 253).<br />
2.3.2. Qualitative versus quantitative approaches to <strong>productivity</strong><br />
<strong>The</strong> approaches to de<strong>fi</strong>ning the nature <strong>of</strong> <strong>productivity</strong> can roughly be divided into<br />
two, qualitative and quantitative. If <strong>productivity</strong> is considered a qualitative<br />
phenomenon, a given word-formation process either has this character or not<br />
(Plag 2006: 538–539). <strong>The</strong> qualitative approach to <strong>productivity</strong> has mainly<br />
concentrated on various kinds <strong>of</strong> phonological, <strong>morphological</strong>, syntactic and<br />
semantic restrictions <strong>of</strong> different word-formation processes, the goal having been<br />
to provide an intensional description <strong>of</strong> all the possible bases <strong>of</strong> a word-formation<br />
process (Evert and Lüdeling 2000: 167). <strong>The</strong> most typical constraints on the<br />
<strong>productivity</strong> <strong>of</strong> word-formation processes are discussed in section 2.3.3.<br />
It has also been claimed that the <strong>productivity</strong> <strong>of</strong> word-formation<br />
processes can be described as a continuum rather than as an all-or-nothing<br />
property in the sense that these processes are productive to a varying extent (Plag<br />
2006: 539). Baayen calls this kind <strong>of</strong> <strong>productivity</strong> scalar, as opposed to an<br />
absolute view <strong>of</strong> <strong>productivity</strong> (1992: 185). Most scholars currently seem to<br />
assume at least some degree <strong>of</strong> scalar <strong>productivity</strong>, whether it means that<br />
<strong>productivity</strong> is in<strong>fi</strong>nitely variable on a scale or that there is a certain number <strong>of</strong><br />
steps on that scale (see, e.g., Bauer 2001: 15–16, 126). Even though the<br />
qualitative and quantitative approaches to <strong>productivity</strong> can be seen as a<br />
dichotomy, the suggestion is that they are nevertheless closely related. As Plag<br />
has put it, “the idea <strong>of</strong> potentiality, which is central to qualitative de<strong>fi</strong>nitions <strong>of</strong><br />
<strong>productivity</strong>, can be expressed in the statistical terms <strong>of</strong> probability” (1999: 22).<br />
<strong>The</strong> distinction between the availability and pr<strong>of</strong>itability <strong>of</strong><br />
<strong>morphological</strong> processes is quite useful in grasping the difference between<br />
qualitative and quantitative aspects <strong>of</strong> <strong>productivity</strong>. <strong>The</strong> terms come originally<br />
from Corbin, who calls them disponibilité and rentabilité, respectively (1987:<br />
177). 10 <strong>The</strong> availability <strong>of</strong> a word-formation process means whether it can be used<br />
10 <strong>The</strong> English translation is by Carstairs-McCarthy (1992: 37).<br />
22
to form new lexemes, and the pr<strong>of</strong>itability <strong>of</strong> a process shows how many bases the<br />
process can affect (Carstairs-McCarthy 1992: 37). Availability is thus a<br />
qualitative notion, while pr<strong>of</strong>itability can be seen as quantitative (Plag 2006: 122).<br />
Pr<strong>of</strong>itability is <strong>of</strong>ten used as synonym for type frequency, that is, the<br />
number <strong>of</strong> words that have been produced using a certain word-formation process<br />
(the number <strong>of</strong> actual words). In this sense, pr<strong>of</strong>itability is a rather straightforward<br />
concept, since it is possible to test it empirically (Férnandez-Dominguez 2007:<br />
60). Availability, on the other hand, is a trickier concept. Again, the problem lies<br />
in the synchronic versus diachronic dichotomy. Processes that are being used, i.e.,<br />
available, at present, are rather unproblematic, but problems arise with processes<br />
that have ceased to be used productively. Férnandez-Domínguez suggests that<br />
availability corresponds to the paradigmatic axis <strong>of</strong> <strong>productivity</strong> and pr<strong>of</strong>itability<br />
to its syntagmatic axis (2007: 63–64). In other words, availability concerns the<br />
selection between competing suf<strong>fi</strong>xes, such as the noun-forming suf<strong>fi</strong>xes -ance, -<br />
al, or -ion.<br />
2.3.3. Measuring <strong>morphological</strong> <strong>productivity</strong><br />
As was stated above, quantitative approaches to <strong>productivity</strong> <strong>of</strong>ten assume that<br />
there is a way to operationalize the problem and determine the exact degree <strong>of</strong><br />
<strong>productivity</strong> <strong>of</strong> a particular word-formation process. According to this view, the<br />
<strong>productivity</strong> values <strong>of</strong> different processes can also be compared. Productivity as a<br />
quantitative phenomenon is probably mentioned <strong>fi</strong>rst by Bolinger, who de<strong>fi</strong>nes it<br />
as “the statistically determinable readiness with which an element enters into new<br />
combinations” (1948: 18). Since then a number <strong>of</strong> methods to measure<br />
<strong>productivity</strong> have been proposed. <strong>The</strong> most important measures will be discussed<br />
next.<br />
Type frequency has been suggested by several authors as a rather<br />
good measure for gauging <strong>productivity</strong>. Baayen calls this kind <strong>of</strong> measure realized<br />
<strong>productivity</strong> (2009: 901–902) or extent <strong>of</strong> use (1992). Haspelmath uses the term<br />
degree <strong>of</strong> generalization (2002: 109). <strong>The</strong> basic assumption with realized<br />
<strong>productivity</strong> is that a productive <strong>morphological</strong> category has many members.<br />
Thus, it can be measured simply by counting the number <strong>of</strong> different types (that<br />
is, different word forms) within a certain <strong>morphological</strong> category.<br />
23
<strong>The</strong> method is not devoid <strong>of</strong> problems, however. Type frequency<br />
does not tell us anything about the availability <strong>of</strong> a process, i.e., whether the<br />
process can be used to form new words in the future or whether it has been<br />
“saturated” (Bauer 2001: 144–145). For example, the suf<strong>fi</strong>x -ment has a high type<br />
frequency, but there are only a few neologisms with -ment in the OED that have<br />
been recorded in the twentieth century (Haspelmath 2002: 109). In this sense, type<br />
frequency can only be used as an indicator <strong>of</strong> past <strong>productivity</strong> (Cowie and<br />
Dalton-Puffer 2002: 416). In addition, if type frequency is assumed to equal the<br />
degree <strong>of</strong> <strong>productivity</strong> <strong>of</strong> a certain word-formation process, it may lead to<br />
counterintuitive results. <strong>The</strong>re are many classes <strong>of</strong> words that have several<br />
members but that are not productive, such as pronouns.<br />
Type frequency has been used as an indicator <strong>of</strong> <strong>productivity</strong>,<br />
especially in historical studies (as in Dalton-Puffer 1996, Cowie and Dalton-<br />
Puffer 2002), since many statistical methods do not function properly if there is<br />
scarcely data available (see below for discussion). Another method used in<br />
diachronic studies is counting the number <strong>of</strong> neologisms that are attested during a<br />
speci<strong>fi</strong>ed period <strong>of</strong> time (Plag 2006: 541). <strong>The</strong> neologisms can be determined with<br />
the help <strong>of</strong> a good historical dictionary, but the reliability <strong>of</strong> the dictionary may<br />
become an issue. For example, lexicographers may overlook new words that are<br />
formed with a very productive pattern (Haspelmath 2002: 110). Bauer also<br />
mentions the danger <strong>of</strong> confusing the actual process <strong>of</strong> word-formation from the<br />
results <strong>of</strong> the word-formation: in English, for example, there are several complex<br />
words (such as payable) which are not created in English but are loans (2001:<br />
145). <strong>The</strong> advantages and disadvantages <strong>of</strong> dictionaries, as opposed to corpora,<br />
are discussed in section 3.2.<br />
A more re<strong>fi</strong>ned view <strong>of</strong> measuring <strong>productivity</strong> comes from Aron<strong>of</strong>f<br />
(1976), who has proposed that <strong>productivity</strong> should be understood as a ratio <strong>of</strong><br />
actual to possible words: the higher the ratio, the greater the <strong>productivity</strong> <strong>of</strong> a<br />
given word-formation rule. However, it is not a simple task to count the number<br />
<strong>of</strong> possible words with that word-formation process. It could, in theory, be<br />
determined by identifying all the possible restrictions on that pattern, but in<br />
practice there are several unproductive rules in English that do not seem to have<br />
any identi<strong>fi</strong>able restrictions (Haspelmath 2002: 110). In addition, since the<br />
possible bases may themselves be complex words that are formed productively,<br />
24
the set <strong>of</strong> possible words becomes practically in<strong>fi</strong>nite (Haspelmath 2002: 110).<br />
Anshen and Aron<strong>of</strong>f have also found out that for highly productive wordformation<br />
processes, the model gives a rather low <strong>productivity</strong> index, since the<br />
number <strong>of</strong> possible words is, in theory, in<strong>fi</strong>nite (1981: 64). <strong>The</strong> most fundamental<br />
problem with Aron<strong>of</strong>f’s model, though, is how to actually count the number <strong>of</strong><br />
possible words. Baayen (1992, 1993) has attempted to formalize Aron<strong>of</strong>f’s model<br />
by introducing the potentiality measure I, which is the ratio <strong>of</strong> the number <strong>of</strong><br />
possible words S and the number <strong>of</strong> actual words V (the reverse <strong>of</strong> Aron<strong>of</strong>f’s<br />
index) with a given af<strong>fi</strong>x.<br />
<strong>The</strong> methods discussed above can be de<strong>fi</strong>ned as quanti<strong>fi</strong>cations, as<br />
opposed to probabilistic measures (see Plag 1999: 24). Baayen and his<br />
collaborators in particular (see, e.g.,Baayen 1992, 1993, 2009; Baayen and Lieber<br />
1991;Chitashvili and Baayen 1993; Baayen and Renouf 1996) have developed a<br />
number <strong>of</strong> statistical methods to measure <strong>productivity</strong> that make use <strong>of</strong> large textcorpora.<br />
Most <strong>of</strong> the statistical measures to gauge <strong>productivity</strong> rely on the<br />
notion <strong>of</strong> the hapax legomenon. <strong>The</strong> words hapax legomenon constitute a Greek<br />
expression that has been adopted from classical studies, and it literally means<br />
‘said (only) once’ (Carstairs-McCarthy 2002: 96). Hapax legomena (or just<br />
‘hapaxes’) are words that occur only once in a certain corpus (Bauer 2005: 325).<br />
With productive <strong>morphological</strong> processes a large number <strong>of</strong> low<br />
frequency words is to be expected, because our mental lexicon “allows the<br />
decomposition <strong>of</strong> the newly encountered word ... and thus the computation <strong>of</strong> the<br />
meaning on the basis <strong>of</strong> the meaning <strong>of</strong> the parts” (Plag 2006: 542). Chitashvili<br />
and Baayen call this kind <strong>of</strong> phenomenon a Large Number <strong>of</strong> Rare Events<br />
(LNRE) distribution (1993: 57). 11 Unproductive processes, on the other hand, tend<br />
to result in a large number <strong>of</strong> high-frequency words (Plag 2006: 542). In a similar<br />
vein, the fact that a productive word-formation rule is available in the mental<br />
lexicon guarantees that all complex words with that af<strong>fi</strong>x can be produced or<br />
understood, even if they have a very low frequency (Baayen and Renouf 1996:<br />
74). This implies that hapaxes can be said to be a good indicator <strong>of</strong> <strong>productivity</strong>.<br />
11 How this distribution can be further used in the statistical study <strong>of</strong> <strong>productivity</strong> will be discussed<br />
in section 2.3.3.<br />
25
It is not important whether the hapax legomena are familiar to the<br />
speaker or not, but rather the fact that they “represent the rate at which new words<br />
are being coined in the language as a whole” (Bauer 2005: 325). 12 It is also worth<br />
noting that not all hapax legomena are necessarily neologisms – while some<br />
hapaxes are just old or obsolete words, some neologisms may occur more than<br />
once in a corpus or be completely unattested (Szymanek 2005: 430–431). Thus,<br />
hapaxes are not a goal in itself: they merely function as a statistical tool for<br />
estimating <strong>productivity</strong> (Baayen 2009: 905–906).<br />
<strong>The</strong> most famous and widely cited <strong>of</strong> Baayen’s measures also makes<br />
use <strong>of</strong> hapaxes. <strong>The</strong> measure has several names; most <strong>of</strong>ten it is called potential<br />
<strong>productivity</strong> (Baayen 2009: 901–902), <strong>productivity</strong> in the strict sense (Baayen and<br />
Lieber 1991: 821), or the category-conditioned degree <strong>of</strong> <strong>productivity</strong> (Baayen<br />
1993: 200). Baayen and Lieber (1991: 809) have formalized the measurement <strong>of</strong><br />
potential <strong>productivity</strong> as follows:<br />
where n 1 represents the number <strong>of</strong> hapax legomena with a given af<strong>fi</strong>x and N the<br />
number <strong>of</strong> tokens <strong>of</strong> all words with the same af<strong>fi</strong>x. <strong>The</strong> resultant <strong>fi</strong>gure is a<br />
decimal that has a value between 0 and 1. <strong>The</strong> higher the value, the more<br />
productive the process is, and vice versa. For example, if there are 100 tokens<br />
with the pre<strong>fi</strong>x re- in the corpus, and 6 types that occur only once, i.e., are<br />
hapaxes, the P value <strong>of</strong> re- is 0.06. For the sake <strong>of</strong> simplicity, this measure will be<br />
mostly referred to as P.<br />
Baayen and Lieber state that P allows measuring “the rate at which<br />
new types are to be expected to appear when N tokens have been sampled” and<br />
estimating the probability <strong>of</strong> encountering new types (1991: 809). <strong>The</strong>refore, if<br />
the sample is representative enough, “P makes a statement <strong>of</strong> what potentially<br />
12 Cowie and Dalton-Puffer state that the”newness” <strong>of</strong> words is usually de<strong>fi</strong>ned in terms <strong>of</strong><br />
dictionaries:”what is not in a given comprehensive dictionary is taken to be new”. This must not be<br />
confused with the psycholinguistic aspect <strong>of</strong> newness, since a word that is well established in a<br />
dictionary may be new to individual speakers (2002: 419).<br />
26
could have been in the sample but has not been actualized for some reason or<br />
other” (1991: 811). In other words, the measure captures the probability that the<br />
next token in the sample is a hapax: according to Baroni, the proportion <strong>of</strong><br />
hapaxes at the Nth token should be a good enough estimate <strong>of</strong> how likely a word<br />
N + 1 will be a hapax (2008: 818).<br />
Baayen and Renouf state that if the size <strong>of</strong> the corpus is increased,<br />
the number <strong>of</strong> hapaxes increases as well, but not in a linear way (1996: 75). This<br />
is explained by the fact that while the sample is still small, the probability <strong>of</strong><br />
encountering a so far unattested word is relatively high. On the other hand, the<br />
larger the sample is, the more likely it already contains that particular word. Thus,<br />
the measure does not allow the comparison <strong>of</strong> P <strong>fi</strong>gures between different corpora<br />
(Baayen 1993: 191). 13 <strong>The</strong> growth rate <strong>of</strong> the vocabulary and hence the slope <strong>of</strong><br />
the vocabulary growth curve varies at different sample sizes. Vocabulary growth<br />
curves will be discussed in section 2.3.3.<br />
Potential <strong>productivity</strong> is not the only statistical method developed by<br />
Baayen and his colleagues. <strong>The</strong> hapax-conditioned degree <strong>of</strong> <strong>productivity</strong> P* was<br />
introduced in Baayen (1993). Also called expanding <strong>productivity</strong>, it captures the<br />
rate at which a category is “expanding and attracting new members”, and is<br />
calculated by dividing the number <strong>of</strong> hapaxes <strong>of</strong> a given category by the number<br />
<strong>of</strong> all the hapaxes in a corpus (Baayen 2009: 901–902). In other words, P* can be<br />
used to study the contribution <strong>of</strong> a certain word-formation process to the<br />
expansion <strong>of</strong> the vocabulary in general. Baayen suggests that P and P* should be<br />
seen as complementary processes, the former being used to distinguish between<br />
productive (even though this is not simple at all) and unproductive processes and<br />
the latter to rank productive processes (1993: 194).<br />
<strong>The</strong> measures introduced in the work <strong>of</strong> Baayen and his colleagues<br />
are not the only statistical methods that have been developed to gauge<br />
<strong>productivity</strong>. Säily and Suomela (2009) have designed a model that makes use <strong>of</strong><br />
accumulation curves and statistical signi<strong>fi</strong>cance testing. In their method, the<br />
corpus is divided into samples that preserve discourse structure. A growth curve<br />
for a <strong>morphological</strong> category (such as a particular af<strong>fi</strong>x) is then plotted on a<br />
13 This is a speci<strong>fi</strong>c instance <strong>of</strong> a more general problem <strong>of</strong> “variable constants” in lexical statistics<br />
(Tweedie and Baayen 1998), cf. type/token ratio.<br />
27
<strong>fi</strong>gure. 14 A sample is picked randomly, the number <strong>of</strong> types or hapaxes is<br />
calculated, and the sample is plotted on a <strong>fi</strong>gure (see Figure 1). Another (random)<br />
sample is picked and added to the previous one, until the whole corpus has been<br />
sampled. <strong>The</strong> procedure is then repeated, for example, a million times with the<br />
help <strong>of</strong> a suitable computer s<strong>of</strong>tware, and the permutation, i.e., the order <strong>of</strong> the<br />
samples changes every time. <strong>The</strong> result looks something like Figure 2: it is<br />
possible to indicate visually the area covered by a certain percentage <strong>of</strong> the<br />
curves, which helps us to see the extent to which the <strong>productivity</strong> rates <strong>of</strong> a certain<br />
feature in a sub-corpus differ from those <strong>of</strong> the corpus as a whole (Säily 2011:<br />
125).<br />
According to Säily, type accumulation curves are a particularly good<br />
measure in comparing <strong>productivity</strong> between different (e.g., gender-speci<strong>fi</strong>c)<br />
subcorpora (2011: 127, 135). <strong>The</strong>re are two ways to plot a type accumulation<br />
curve: either the number <strong>of</strong> types as a function <strong>of</strong> corpus size, or the number <strong>of</strong><br />
hapaxes as a function <strong>of</strong> the number <strong>of</strong> types. Hapax accumulation curves,<br />
however, require a large corpus in order to produce statistically valuable data in<br />
which the results are not skewed (Säily 2011).<br />
14 <strong>The</strong> curves can be plotted either for types (as a function <strong>of</strong> token frequencies) or hapaxes (as a<br />
function <strong>of</strong> running words in the corpus) (see Säily 2011: 126).<br />
28
Figure 1.Two type accumulation curves. Each tick mark represents the addition<br />
<strong>of</strong> one sample. From Säily and Suomela (2009: 102).<br />
Figure 2.Bounds for 1,000,000 accumulation curves for the suf<strong>fi</strong>x –ity in the 17 th -<br />
century part <strong>of</strong> the CEEC.Gender-based subcorpora plotted on the curves, which<br />
show that women have a signi<strong>fi</strong>cantly lower type frequency. From Säily and<br />
Suomela (2009: 100, as cited in Säily 2011: 126).<br />
29
2.3.3. Lexical statistics and <strong>productivity</strong><br />
As was shown with the example <strong>of</strong> type accumulation curves, the study <strong>of</strong><br />
<strong>productivity</strong> can bene<strong>fi</strong>t from what is usually called lexical statistics, i.e., the<br />
study <strong>of</strong> word frequency distributions and the probabilities <strong>of</strong> occurrence <strong>of</strong><br />
different types across texts and corpora (see Baroni 2008: 803). How different<br />
word types are distributed in texts usually follows some pattern. For example, it<br />
was pointed out above that productive word-formation processes are characterized<br />
by a large number <strong>of</strong> low-frequency words, such as hapax legomena or dis<br />
legomena (words that occur twice in a corpus). Interestingly, samples consisting<br />
<strong>of</strong> naturally occurring text, independently <strong>of</strong> size, language, or textual typology,<br />
are characterized by a similar pattern <strong>of</strong> few very frequent words (typically<br />
function words such as a, the, <strong>of</strong>, etc.) and a long tail <strong>of</strong> low-frequency types<br />
(Baroni 2008: 810). A statistical model like this is called a “large number <strong>of</strong> rare<br />
events” (LNRE) distribution (Baayen 2001: 51).<br />
Such a skewed distribution with “few giants ... and an army <strong>of</strong><br />
dwarves” (Baroni 2008: 814) can be predicted by Zipf’s law (Zipf 1949, 1965). 15<br />
<strong>The</strong> model shows how frequency is a non-linearily decreasing function <strong>of</strong> the rank<br />
<strong>of</strong> a word in the text. For example, the most frequent word in the text has the rank<br />
1, the second most frequent word has 2, and so on. Zipf’s law is formalized as<br />
follows:<br />
( )<br />
( )<br />
where f(w) is the frequency <strong>of</strong> a word w, and r(w) the rank <strong>of</strong> the same word,<br />
while C and a are constants that depend on the corpus (Baroni 2008: 813). If a is<br />
assumed to be 1, then C is the frequency <strong>of</strong> the most frequent word in a corpus<br />
(having thus rank 1). If the most frequent word occurs 3000 times in the corpus,<br />
then C = 3000. <strong>The</strong> second most frequent word then has a frequency <strong>of</strong> C/2 =<br />
1500, the 100 th most frequent word C/100 = 3, and so on. Thus the frequencies <strong>of</strong><br />
words decrease rapidly among the high ranks, but slows down as the rank<br />
decreases.<br />
15 Due to the limited scope <strong>of</strong> the study, the deep mathematics underlying the Zipf-Mandelbrot<br />
model are discussed here only cursorily. A more thorough account <strong>of</strong> the mathematical motivation<br />
<strong>of</strong> the model is given by Baayen (2001).<br />
30
Even though Zipf’s law is sound in theory, it does not provide a<br />
perfect <strong>fi</strong>t for expected values <strong>of</strong> word frequencies (Baroni 2008: 815). A better <strong>fi</strong>t<br />
is obtained when the extra parameter b is added to the formula:<br />
( )<br />
( ( ) )<br />
<strong>The</strong> resulting formula is called Zipf-Mandelbrot’s law, <strong>of</strong> which the original<br />
Zipf’s law is a special case with b = 0 (Baroni 2008: 815). <strong>The</strong> optimal parameters<br />
can be determined with the help <strong>of</strong> a computer. <strong>The</strong> zipfR package, available for<br />
the R s<strong>of</strong>tware, has been used in this study (see Evert and Baroni 2007).<br />
Table 1. A toy rank/frequency pr<strong>of</strong>ile (left panel) and frequency spectrum (right<br />
panel)<br />
r f f V(f)<br />
1 13 1 10<br />
2 7 2 5<br />
3 5 3 3<br />
4 2 4 2<br />
5 1 5 1<br />
Baroni mentions two useful tools that help capture word frequency<br />
distributions in a visual way: rank/frequency pr<strong>of</strong>iles and frequency spectra<br />
(2008: 806). A rank/frequency pr<strong>of</strong>ile resembles a type frequency list on which<br />
the types in a corpus and their respective frequencies are placed in a descending<br />
order. In a rank/frequency pr<strong>of</strong>ile, the type labels are replaced by their frequencybased<br />
ranks, 1 referring to the most frequent type and so on. A frequency<br />
spectrum, on the other hand, reports how many types have a certain frequency.<br />
Table 1 <strong>of</strong>fers an example <strong>of</strong> a rank/frequency pr<strong>of</strong>ile and a frequency spectrum<br />
31
which can be seen as reverse sides <strong>of</strong> the same phenomenon. A more visual way<br />
to illustrate those two frequency distributions is to present them as a bar plot<br />
graph. An example is <strong>of</strong>fered in <strong>fi</strong>gure 3, which draws from data for the present<br />
study.<br />
Figure 3. A rank/frequency pr<strong>of</strong>ile and a frequency spectrum for the initial<br />
combining form hyper- in the written part <strong>of</strong> the BNC.<br />
What makes rank/frequency pr<strong>of</strong>iles and frequency spectra<br />
interesting from the point <strong>of</strong> view <strong>of</strong> <strong>productivity</strong> is that if we know the<br />
distribution <strong>of</strong> type probabilities <strong>of</strong> a certain word-formation process, the expected<br />
vocabulary size V and the number <strong>of</strong> hapax legomena V 1 can be calculated for any<br />
sample size N (Baayen 2001: 41–51). Since potential <strong>productivity</strong> P essentially<br />
measures the vocabulary growth rate <strong>of</strong> different processes and consequently the<br />
slope <strong>of</strong> the growth curves for that process, comparing the slopes will give us<br />
information about the differences in <strong>productivity</strong> between different processes. <strong>The</strong><br />
vocabulary growth <strong>of</strong> a process can be visualized by vocabulary growth curves,<br />
which are essentially the same thing as the type accumulation curves (Säily and<br />
Suomela 2009, Säily 2011), displaying the type frequency V <strong>of</strong> a certain wordformation<br />
process as the function <strong>of</strong> its token frequency N. An exemplary<br />
vocabulary growth curve is presented in Figure 4.<br />
32
Figure 4. Vocabulary growth curves for the Italian pre<strong>fi</strong>xes ri- (interpolated) and<br />
ultra- (extrapolated). <strong>The</strong> x axis displays the corpus size N and the y axis the<br />
number <strong>of</strong> types V in relation to N. From Evert and Baroni 2007.<br />
However, since the vocabulary does not grow linearily, direct<br />
comparison <strong>of</strong> differently sizes samples, i.e., different word-formation processes,<br />
is not reliable. In the case <strong>of</strong> samples <strong>of</strong> different sizes, it is therefore necessary to<br />
extrapolate the growth curve <strong>of</strong> the smaller sample and predict the growth rate for<br />
an arbitrary sample size. This involves computing the expected sample size for a<br />
smaller corpus in order to “stretch” it to the length <strong>of</strong> a larger corpus (Baroni<br />
2008: 820). This can be done with the zipfR package for the R statistics s<strong>of</strong>tware.<br />
16<br />
Figure 4 (from Evert and Baroni 2007) illustrates the difference<br />
between the <strong>productivity</strong> <strong>of</strong> two Italian pre<strong>fi</strong>xes. <strong>The</strong> pre<strong>fi</strong>x ultra- has a smaller<br />
population that the pre<strong>fi</strong>x ri-, but when the sample is extrapolated to equal the size<br />
16 For a more detailed account <strong>of</strong> the zipfR package and its implementation, see Evert and Baroni<br />
2007.<br />
33
<strong>of</strong> ri-, and the resulting vocabulary growth curve is compared to that <strong>of</strong> ri-, it can<br />
be seen that it in fact has a steeper curve. This tells us that it has a higher growth<br />
rate, i.e., a higher P value. In general, it is typical for a vocabulary growth curve<br />
to grow more steeply at lower sample sizes, because if a corpus is small, it is more<br />
likely to encounter new, unattested types. As the corpus grows, the likelihood<br />
decreases.<br />
2.3.4. <strong>The</strong> psycholinguistic aspect <strong>of</strong> <strong>productivity</strong><br />
Attempts to measure degrees <strong>of</strong> <strong>productivity</strong> constitute only one aspect in the<br />
study <strong>of</strong> this phenomenon. Another issue is how speakers process, storage,<br />
retrieve and produce <strong>morphological</strong>ly complex words in terms <strong>of</strong> their mental<br />
lexicon (cf. Bauer 2001: 101, 112), and what the role <strong>of</strong> <strong>productivity</strong> is in this<br />
schema. <strong>The</strong> primary interest <strong>of</strong> this study is not the psycholinguistic aspect <strong>of</strong><br />
<strong>productivity</strong>, but it is worth discussing nevertheless, since many quantitative<br />
measures <strong>of</strong> <strong>productivity</strong> are also psycholinguistically motivated, as will be shown<br />
below.<br />
A crucial question at this point is whether complex words are stored<br />
as single units or as distinct morphemes in the mental lexicon, while the latter way<br />
involving <strong>morphological</strong> parsing. 17 <strong>The</strong> former way is more adequate from the<br />
point <strong>of</strong> view <strong>of</strong> processing, the latter facilitates the storage <strong>of</strong> the forms<br />
(Frauenfelder and Schreuder 1992: 166–167). In Frauenfelder and Schreuder’s<br />
model, each complex word is simultaneously processed in both ways, the faster<br />
route winning the race (hence the term <strong>morphological</strong> race model) (1992: 175).<br />
Factors that influence which route will win include token frequency (high token<br />
frequency favours the direct route) and transparency (1992: 176–177). <strong>The</strong> model<br />
is related to Baayen’s statistical <strong>productivity</strong> measures by predicting that the<br />
parsing route wins with transparent, infrequent words – the very properties <strong>of</strong><br />
words that are assumed to be <strong>of</strong> high <strong>productivity</strong> in Baayen’s model.<br />
Relying on Frauenfelder and Schreuder’s work, Baayen (1993)<br />
discusses the activation level A, a measure that attempts to determine which route<br />
17 Parsing has several different (although related) senses. In this context, by parsing one means the<br />
way speakers analyse complex words into their constituent parts (see, e.g., Frauenfelder and<br />
Schreuder 1992: 175).<br />
34
to choose in the race. <strong>The</strong> activation level <strong>of</strong> a word-formation process is<br />
formalized as the number <strong>of</strong> tokens containing that process with a frequency less<br />
than a threshold θ (1993: 196). If the frequency is higher than the threshold, the<br />
word is processed by the direct route, and while words with a lower frequency are<br />
parsed (1993: 196). For example, in the present data, the activation level for<br />
pseudo- would be 491, if θ = 8 (the same as Baayen uses). Thus, all types with a<br />
frequency less than 8 are parsed. Baayen does not tell us how exactly the<br />
threshold θ should be determined, but he writes that if the value <strong>of</strong> θ is high, more<br />
opaque forms will be included in the frequency counts (1993: 203).<br />
Baayen’s model is further improved by Hay (2001, 2003), who<br />
states that instead <strong>of</strong> the absolute frequency <strong>of</strong> a word, the degree <strong>of</strong><br />
decomposability <strong>of</strong> complex words depends on the relative frequency <strong>of</strong> the<br />
derived word and the base, i.e., the frequency <strong>of</strong> the derived word divided by the<br />
frequency <strong>of</strong> its base. For example, the derivative business is more frequent than<br />
its base busy, which means that it has a high relative frequency and that it is<br />
accessed as whole (Plag 2006: 548). By contrast, words with a low relative<br />
frequency are <strong>of</strong>ten <strong>morphological</strong>ly productive and semantically transparent.<br />
Plag summarizes that the semantical transparency <strong>of</strong> productive word-formation<br />
rules is the consequence <strong>of</strong> cognitive processing (2006: 548). Hay and Baayen<br />
(2002) attempt to further re<strong>fi</strong>ne the model by de<strong>fi</strong>ning the speci<strong>fi</strong>c parsing line<br />
above which derived words are most likely to be parsed.<br />
Neither Baayen’s activation level nor Hay’s relative frequency<br />
model will be applied in this study as such. However, psycholinguistic evidence<br />
might prove useful in determining the appropriate units <strong>of</strong> analysis, i.e., how to<br />
deal with pseudo-af<strong>fi</strong>xed words. Baayen actually <strong>of</strong>fers some guidelines on how<br />
to determine appropriate units <strong>of</strong> analysis when studying <strong>productivity</strong> (1993: 200–<br />
201). Questions like this will be further discussed in section 4.<br />
2.3.4. Constraints on <strong>productivity</strong><br />
A word-formation pattern cannot apply to any word in a language, but only to a<br />
certain subset <strong>of</strong> words (Rainer 2005: 335). In other words, there are various<br />
factors that restrict the potential combinations <strong>of</strong> af<strong>fi</strong>xes. It has even been<br />
proposed that all <strong>morphological</strong> rules are equally productive, so that the<br />
35
differences in (apparent) <strong>productivity</strong> are caused by restrictions on the rules.<br />
However, there are many rules that do not seem to be restricted in any general<br />
way, which corroborates this idea. 18 <strong>The</strong> study <strong>of</strong> the various restrictions on wordformation<br />
processes is related to the qualitative aspect <strong>of</strong> <strong>productivity</strong>.<br />
<strong>The</strong> subset <strong>of</strong> possible bases for a word-formation process is called a<br />
domain (Rainer 2005: 335). This phenomenon can be called a restriction, but most<br />
scholars use the term constraint instead. Bauer, for example, justi<strong>fi</strong>es the use <strong>of</strong><br />
constraint by the fact that the restrictions are not always absolute (2001: 126).<br />
<strong>The</strong>re are several ways to classify different kinds <strong>of</strong> constraints. Rainer, for<br />
example, divides them into universal and language-speci<strong>fi</strong>c (2005: 335). <strong>The</strong><br />
former apply to all languages, while the latter depend on the language in question.<br />
Another division <strong>of</strong> constraints is <strong>of</strong>ten made between external<br />
factors (factors that are mainly related to fashion, knowledge, and attitudes) and<br />
internal factors (caused by the internal structure <strong>of</strong> a language) (Férnandez-<br />
Domínguez 2007: 69). I will mainly follow Rainer’s model, although my focus<br />
will be on the language-speci<strong>fi</strong>c constraint relevant for the English language and<br />
especially on the various structural constraints that comprise what Férnandez-<br />
Domínguez calls “all the levels <strong>of</strong> linguistic description” (2007: 69). All the<br />
structural constraints (i.e., phonological, <strong>morphological</strong>, syntactic, semantic,<br />
pragmatic) are thus language-speci<strong>fi</strong>c, and the only constraints treated here that<br />
fall under the category <strong>of</strong> universal constraints are the various types <strong>of</strong> blocking.<br />
Phonological constraints on <strong>morphological</strong> compatibility are a very<br />
common type, illustrated by phonologically conditioned allomorphs, such as the<br />
complementary set <strong>of</strong> plural markers /iz/, /s/ and /z/ (Bauer 2001: 128). Bauer<br />
mentions three properties <strong>of</strong> the base that may constrain what af<strong>fi</strong>xes may attach<br />
to it (2001: 129–130). Sometimes the segmental make-up <strong>of</strong> the base is important:<br />
the verb-forming -en only attaches to bases that end in a stop or a fricative). In<br />
other cases the suprasegmental make-up, i.e., stress, is what matters: the nounforming<br />
suf<strong>fi</strong>x -al is only attached to bases that have the stress on the <strong>fi</strong>rst<br />
syllable. Sometimes the number <strong>of</strong> syllables in the base determines which af<strong>fi</strong>xes<br />
may attach to it. As regards the third case, Bauer does not mention any examples<br />
18 For discussion, see Haspelmath (2002: 111).<br />
36
in English. Phonological restrictions are a common example <strong>of</strong> language-speci<strong>fi</strong>c<br />
constraints.<br />
With <strong>morphological</strong> constraints on <strong>productivity</strong>, the <strong>morphological</strong><br />
structure <strong>of</strong> the base determines what kinds <strong>of</strong> af<strong>fi</strong>xes may attach to it (Bauer<br />
2001: 130). Again, Bauer lists three options (2001: 130–132). Some af<strong>fi</strong>xes apply<br />
only to a certain <strong>morphological</strong>ly de<strong>fi</strong>ned set <strong>of</strong> bases. <strong>The</strong> set can be de<strong>fi</strong>ned in<br />
terms <strong>of</strong> etymology, for example. This is <strong>of</strong>ten called level ordering (see below<br />
for discussion). Sometimes a certain af<strong>fi</strong>x may be added only to a derived or to an<br />
underived base. <strong>The</strong> base may also have to end in a particular af<strong>fi</strong>x before any<br />
further af<strong>fi</strong>xes can be added. In general, pre<strong>fi</strong>xes have fewer word-class<br />
restrictions than suf<strong>fi</strong>xes, because they do not usually alter the word-class <strong>of</strong> the<br />
base they attach to (Nevalainen 1999: 355).<br />
<strong>The</strong> distinction between <strong>morphological</strong> and syntactic constraints is<br />
not always clear-cut. Bauer draws the line by de<strong>fi</strong>ning <strong>morphological</strong> constraints<br />
as ones that concern the internal structure <strong>of</strong> the word, while syntactic constraints<br />
deal with the way a word is used in context (2001: 133). A typical example <strong>of</strong> a<br />
syntactic constraint is the fact that many af<strong>fi</strong>xes can only be added to a certain<br />
word class as a base (Bauer 2001: 133). Sometimes the transitivity <strong>of</strong> the verb<br />
base is <strong>of</strong> importance: for example, the suf<strong>fi</strong>x -able may only be used on transitive<br />
verbs.<br />
Level ordering is one <strong>of</strong> the language-speci<strong>fi</strong>c constraints on<br />
<strong>productivity</strong>. Level ordering arises because many languages have several<br />
<strong>morphological</strong> strata, a native and a non-native one, which both have some<br />
combinatory restrictions (Rainer 2005: 340). For example, in English, many<br />
Latinate af<strong>fi</strong>xes are restricted to Latinate bases only (Haspelmath 2002: 108).<br />
However, since speakers do not necessarily know whether a stem belongs to the<br />
native or non-native stratum, the restriction on the basis <strong>of</strong> <strong>morphological</strong> strata is<br />
sometimes unstable: the suf<strong>fi</strong>x -ous has been added to some native bases as well<br />
(murderous, thunderous) (Haspelmath 2002: 108).<br />
Finally, pragmatic and sociolinguistic factors may also restrict the<br />
<strong>productivity</strong> <strong>of</strong> word-formation processes. While semantic constraints depend on<br />
the linguistic nature <strong>of</strong> the base, pragmatic constraints (as opposed to structural<br />
constraints) involve the way in which words are actually used (Bauer 2001: 135).<br />
Certain patterns are only used in particular situations, and it requires language<br />
37
competence to know in which situations to use which patterns (Rainer 2005: 349).<br />
One <strong>of</strong> the clearest cases <strong>of</strong> pragmatic restrictions is fashion and the “extralinguistic<br />
developments in the society that make certain words or <strong>morphological</strong><br />
elements desirable to use” (Plag 1999: 39).<br />
Another pragmatic constraint is the general need for a certain word<br />
(Plag 1999: 39). <strong>The</strong> need may either relate to the naming or labelling <strong>of</strong> a new<br />
concept or entity, or to the speakers’ need to condense information due to stylistic<br />
purposes or text cohesion. Yet another, a rather closely related restriction is the<br />
nameability requirement: all new lexemes must denote something nameable in the<br />
language (Plag 1999: 40).<br />
Semantic constraints are not very common, and are mainly<br />
concerned with cases where the referent <strong>of</strong> the base must follow certain semantic<br />
requirements. For example, the referent <strong>of</strong> the nouns with the suf<strong>fi</strong>x -ee must be<br />
sentient (Barker 1998; cited in Rainer 2005: 349). Semantic restrictions are a<br />
common type <strong>of</strong> constraint, semantic blocking being one example. Carstairs-<br />
McCarthy de<strong>fi</strong>nes semantic blocking as a situation in which “the existence <strong>of</strong> a<br />
word (...) with a particular meaning inhibits the <strong>morphological</strong> derivation, even by<br />
formally regular means, <strong>of</strong> another word with precisely that meaning” (2002: 91).<br />
His example makes use <strong>of</strong> the adjectives curious and glorious. A formally regular<br />
way to form nouns from these adjectives would be to employ the suf<strong>fi</strong>x -ity. With<br />
curiosity, this seems to work, the result being curiosity. However, with glorious,<br />
the corresponding noun is glory, not *gloriosity: the existence <strong>of</strong> gloriosity is thus<br />
blocked by the word glory. Even a formally regular word-formation process can<br />
be blocked (ibid.). One reason for the existence <strong>of</strong> the phenomenon <strong>of</strong> semantic<br />
blocking is the tendency to avoid exact synonymy in most languages (Carstairs-<br />
McCarthy 2002: 85).<br />
Rainer calls this kind <strong>of</strong> blocking token blocking, and writes that<br />
children and non-native speakers <strong>of</strong>ten produce formations like gloriosity, which<br />
indicates that the “existence” <strong>of</strong> a synonymous word refers to the mental lexicon<br />
<strong>of</strong> the speaker, not to language as a social institution (2005: 336–337). Besides<br />
token blocking, Rainer mentions type blocking, in which case the unacceptability<br />
<strong>of</strong> a complex word is due to a synonymous pattern that takes precedence (2005:<br />
337–338). He illustrates type blocking by the set <strong>of</strong> rival suf<strong>fi</strong>xes in German,<br />
namely -heit, -ität, and -ie (2005: 338). <strong>The</strong> suf<strong>fi</strong>x -heit is used with adjectives<br />
38
with <strong>fi</strong>nal stress, and it is fully productive. However, its domain is diminished by -<br />
ität (with learned bases) and, more rarely, -ie (with adjectives ending in -phil).<br />
Férnandez-Domínguez de<strong>fi</strong>nes both type blocking and token blocking as<br />
examples <strong>of</strong> synonymy blocking, as opposed to homonymy blocking, which<br />
prevents the coining <strong>of</strong> a new word because <strong>of</strong> structural overlapping with an<br />
existing word (2007: 71–73). For example, the word liver ‘inner organ’ prevents<br />
the coining <strong>of</strong> *liver ‘someone who lives’ (Plag 1999: 50; cited in Férnandez-<br />
Domínguez 2007: 73).<br />
To conclude, it can be said that new words are always formed for a<br />
speci<strong>fi</strong>c need, and thus constraints may not tell us everything about the<br />
pr<strong>of</strong>itability <strong>of</strong> word-formation processes (Bauer 2001: 143). Férnandez-<br />
Domínguez goes as far as to state that the naming need on the part <strong>of</strong> the language<br />
community is so strong that it may even “knock out” the restrictions caused by<br />
various constraints (2007: 84). In any case, the application <strong>of</strong> different wordformation<br />
processes is seldom straightforward, and there are several exceptions to<br />
the rules.<br />
3. Material and methods<br />
3.1. A closer look at the <strong>selected</strong> combining forms<br />
As was stated earlier, combining forms can be divided into initial and <strong>fi</strong>nal<br />
combining forms. In this study, I will compare the two and try to <strong>fi</strong>nd out whether<br />
they differ in <strong>productivity</strong>. To do so, I will compare the values <strong>of</strong> Baayen’s<br />
potential <strong>productivity</strong> P <strong>of</strong> three initial combining forms (hyper-, ultra-, and<br />
pseudo) as well as three <strong>fi</strong>nal combining forms (-logy, -graphy, and -nomy), in<br />
order to determine whether there are any major differences in the <strong>productivity</strong><br />
rates between these two subtypes. <strong>The</strong>se six combining forms were chosen<br />
because they are among the most frequent ones in the BNC. Sample size is<br />
important in performing statistical analyses on the combining forms, since many<br />
<strong>of</strong> the measures used in this study do not work properly if the sample is too small<br />
(see section 2.3.3. for discussion).<br />
39
It must be noted that some neoclassical elements have two kinds <strong>of</strong><br />
combining forms (with allomorphic variation) derived from the same lexical<br />
word, those that can occur in both initial and <strong>fi</strong>nal position (cf. Fradin 2000: 13).<br />
For example the Greek λόγο-ς ’word, speech’ occurs in words like logopedics and<br />
biology. A similar example is the Greek form -γράϕος, which is used in<br />
formations like graphology and biography.<br />
Unless otherwise indicated, the OED constitutes the source in the<br />
following description <strong>of</strong> the <strong>selected</strong> combining forms. <strong>The</strong> problem with the<br />
OED, however, is that even though some <strong>of</strong> the <strong>selected</strong> combining forms, the<br />
<strong>fi</strong>nal ones in particular, have rather sketchy entries. McCauley writes that the<br />
amount <strong>of</strong> information in each entry is to some extent determined by the<br />
frequency, longevity, and complexity <strong>of</strong> the word or element in the English<br />
language (2006: 99). In addition, most entries are rather old and date back to the<br />
late 19 th century (see section 3.2. for discussion). However, the OED is currently<br />
undergoing a thorough revision that will also affect entries for combining forms:<br />
for example, in the second edition <strong>of</strong> the OED, there is sometimes ambiguity in<br />
whether a certain element is an af<strong>fi</strong>x or a combining form (McCauley 2006: 97).<br />
hyper-<br />
Hyper- is <strong>of</strong>ten classi<strong>fi</strong>ed as a pre<strong>fi</strong>x instead <strong>of</strong> a combining form, probably<br />
because it has its origin in the Greek preposition (ὑπέρ ‘over, beyond, over much,<br />
above measure’) instead <strong>of</strong> a lexical word. Nevalainen classi<strong>fi</strong>es hyper- as an<br />
intensifying pre<strong>fi</strong>x, and writes that it is the Greek cognate <strong>of</strong> Latin super-, having<br />
the meaning <strong>of</strong> ‘over, too much’ (1999: 388). <strong>The</strong> OED states that hyper- has been<br />
extensively used in English since the 17 th century in the formation <strong>of</strong> new<br />
compounds (<strong>of</strong>ten on Greek analogies), and that it combines relatively freely with<br />
nouns and adjectives.<br />
pseudo-<br />
40
Pseudo- has its origins in the Greek root ψευδο- ‘false, falsity, falsehood’, and is<br />
also quite frequently attached to Latin bases as well. <strong>The</strong> <strong>fi</strong>rst attestations <strong>of</strong><br />
formations with pseudo- in English date back to early 15 th century in a religious<br />
context, but it did not become common until the 17 th century. <strong>The</strong> <strong>fi</strong>rst native<br />
formations (e.g., pseudo-prophetical) date back to that period as well. From the<br />
19 th century onwards, pseudo- combined with nouns and adjectives becomes<br />
frequent in scienti<strong>fi</strong>c terminology in particular.<br />
ultra-<br />
Unlike the majority <strong>of</strong> the combining forms in this study, ultra- has its origins in<br />
Latin instead <strong>of</strong> Greek. It is derived from a Latin preposition that has a meaning<br />
‘beyond’, and like hyper-, it is <strong>of</strong>ten classi<strong>fi</strong>ed as a measurement pre<strong>fi</strong>x as well.<br />
<strong>The</strong> original meaning in English is ‘lying spatially beyond or on the other side <strong>of</strong>’<br />
(as in ultra-terrestrial, ultra-zodiacal), or ‘going beyond, transcending the limits<br />
<strong>of</strong>’ (ultra-human, ultra-rational). However, from the early 19 th century onwards, a<br />
third sense, ‘an excessive or extreme quality <strong>of</strong> an adjective’ (ultra-cautious,<br />
ultra-cosmopolitan) has steadily become more common (Hughes 2000: 346–347).<br />
-logy<br />
<strong>The</strong> origin <strong>of</strong> -logy is the Greek element λόγο-ς ’word, speech’ or the<br />
corresponding root λόγ-. 19 Derivatives <strong>of</strong> logy have two distinct senses in English.<br />
<strong>The</strong> former group includes words with the sense ‘saying or speaking’ (eulogy,<br />
necrology), while the latter sense can be seen in the names <strong>of</strong> sciences or<br />
disciplines (biology, astrology). All the modern formations <strong>of</strong> -logy can be said to<br />
have a corresponding formations using the combining forms -logical and -logist<br />
(called suf<strong>fi</strong>xal expansion by Prćić (2007: 386–387). Kastovsky states that -logy is<br />
actually much more suf<strong>fi</strong>x-like than lexeme-like (2009: 5).<br />
19 Sometimes the thematic vowel -o- is included by dictionary-makers, which results in the form -<br />
ology in English <strong>The</strong> status <strong>of</strong> the linking element -o- is discussed in section 2.2.2.<br />
41
-graphy<br />
<strong>The</strong> form -graphy is closely related to the variant -graph. However, formations<br />
with -graph are not included in this study. <strong>The</strong> combining form -graphy is derived<br />
from the Greek –γράϕος, which denotes writing. Often the words that have the<br />
component -graphy refer to sciences (geography, bibliography), but they can also<br />
denote writing, drawing, or graphic representation (calligraphy, lithography,<br />
photography). Hybrid formations with -graphy are rare, stratigraphy being one <strong>of</strong><br />
the few examples. As with -logy, words with -graphy can have the corresponding<br />
derivatives -graphic and -graphical.<br />
-nomy<br />
<strong>The</strong> combining form -nomy is derived from the ancient Greek element –νομία, a<br />
variant <strong>of</strong> the base νέμειν ‘to deal, distribute, hold, manage’. <strong>The</strong> earliest<br />
attestations <strong>of</strong> -nomy are astronomy and economy (both from the 16 th century).<br />
<strong>The</strong>re is also a small number <strong>of</strong> words (from the 16 th century onwards) that denote<br />
‘law’, ‘rule’, or ‘government’, as in autonomy.<br />
3.2. Corpora versus dictionaries in the study <strong>of</strong> <strong>productivity</strong><br />
It was stated above that the most widely-used methods to measure <strong>productivity</strong> are<br />
either corpus-based or dictionary-based. Both types <strong>of</strong> methods have their<br />
advantages and disadvantages. Corpora are generally considered to be a more<br />
representative sample <strong>of</strong> language than dictionaries, but the issue is not so<br />
straightforward, as will be shown.<br />
Historical dictionaries in particular are <strong>of</strong>ten partial or even<br />
incomplete, even though they are useful in providing information about the early<br />
attestations <strong>of</strong> lexemes (Palmer 2009: 8). Nevalainen mentions the various<br />
problems with the OED as regards chronological study <strong>of</strong> words. For example,<br />
issues such as chronological coverage are <strong>of</strong>ten ignored (1999: 337). <strong>The</strong>refore,<br />
the citations in the OED do not necessarily reflect the full variety <strong>of</strong> texts in a<br />
42
given period. For the earliest stages <strong>of</strong> language, the sampling is <strong>of</strong>ten sparse, and<br />
a word may have been in use long before its <strong>fi</strong>rst written record (Baayen 2008:<br />
909–910). In a similar vein, Schäfer writes that the OED reflects the frequency<br />
and status <strong>of</strong> a word at the moment <strong>of</strong> compilation <strong>of</strong> the dictionary instead <strong>of</strong> the<br />
period <strong>of</strong> origin (1989: 69).<br />
According to Baayen and Lieber, some <strong>of</strong> the bene<strong>fi</strong>ts <strong>of</strong> corpora as<br />
opposed to dictionaries are that they reflect actual language usage (1991: 803).<br />
Indeed, as Baayen and Renouf point out, “it is commercially unattractive to print<br />
thousands <strong>of</strong> words the meaning <strong>of</strong> which is immediately clear to anyone familiar<br />
with the basic meaning <strong>of</strong> productive af<strong>fi</strong>xes” (1996: 69). According to Booij,<br />
dictionaries only register which new complex words have been established in the<br />
lexicon, while true <strong>morphological</strong> <strong>productivity</strong> is reflected by “complex words<br />
that never make it into dictionary”, i.e., hapaxes (2005: 69). In a similar vein,<br />
Baayen and Renouf state that dictionary-based methods may underestimate the<br />
<strong>productivity</strong> rates <strong>of</strong> af<strong>fi</strong>xes (1996: 70). In addition, the inclusion <strong>of</strong> a word into a<br />
dictionary is always arbitrary to some extent, which means that a dictionary<br />
cannot be considered as a representative sample <strong>of</strong> a language (Plag 1999: 27).<br />
<strong>The</strong> size <strong>of</strong> the corpus is an important factor in all kinds <strong>of</strong> linguistic<br />
analysis, and in the study <strong>of</strong> <strong>productivity</strong> in particular: in a small corpus, most<br />
hapax legomena will be well-known words <strong>of</strong> the language, but if the corpus size<br />
increases, the proportion <strong>of</strong> neologisms among the hapaxes increases as well (Plag<br />
2006: 542–543; see also Baayen and Lieber 1991: 813). For unproductive<br />
processes, increasing the corpus size does not add new types after a certain<br />
“saturation point” has been reached (Lüdeling et al. 2000: 58). In this vein, the use<br />
<strong>of</strong> the BNC is justi<strong>fi</strong>ed, since it is one <strong>of</strong> the largest English corpora available.<br />
Other corpora generally considered large enough include the COBUILD / Bank <strong>of</strong><br />
English corpus and the COCA corpus <strong>of</strong> contemporary American English (see<br />
Plag 1999: 25–26). Historical corpora, on the other hand, are <strong>of</strong>ten too small for<br />
studying potential <strong>productivity</strong> P (Palmer 2009: 8–9).<br />
<strong>The</strong> data for this study was obtained from the written part <strong>of</strong> the <strong>The</strong><br />
British National Corpus (BNC). In total, the BNC comprises just under 100<br />
million word tokens <strong>of</strong> which roughly 89 million words represent written British<br />
English that falls into various genres, or registers, such as <strong>fi</strong>ction, academic prose,<br />
and personal letters (see Burnard 2007). One <strong>of</strong> the possible disadvantages with<br />
43
the BNC is the fact that since it does not cover the last twenty years or so, the<br />
most recent changes or trends in the English word-formation are not covered.<br />
However, in this study, the use <strong>of</strong> the BNC is justi<strong>fi</strong>ed, because it has already been<br />
used in many studies on the <strong>productivity</strong> <strong>of</strong> various af<strong>fi</strong>xes. This enables the<br />
comparison <strong>of</strong> the <strong>productivity</strong> rates to the ones calculated for the <strong>selected</strong><br />
combining forms. When necessary, and especially to provide supplementary<br />
information on the behaviour <strong>of</strong> these combining forms, the Corpus <strong>of</strong><br />
Contemporary American English (COCA) will also be consulted, especially in the<br />
discussion section.<br />
It must be noted, however, that a corpus-based <strong>productivity</strong> measure<br />
is not devoid <strong>of</strong> problems, either. Lüdeling et al. criticize Baayen for assuming<br />
that all electronic corpora are automatically perfectly prepared (2000: 59). <strong>The</strong>re<br />
are, however, at least three corpus pre-processing problems that must be taken<br />
into consideration: mistagged items, typographical errors, and repetition <strong>of</strong> the<br />
same texts in corpus composition. Similar problems are even more common in<br />
historical data, since historical corpora <strong>of</strong>ten contain several spelling variants <strong>of</strong><br />
the same af<strong>fi</strong>xes (see Säily and Suomela 2009: 93). Tagging is not a problem in<br />
the present study, since we are only interested in the internal make-up <strong>of</strong> lexemes.<br />
Even though the BNC has been claimed to have some inherent structural<br />
problems, such as wrongly classi<strong>fi</strong>ed texts and overly broad categories (see Lee<br />
2001), the sampling has been done carefully (see Burnard 2007: section 1.4.4 for<br />
the selection procedures employed), so that the repetition <strong>of</strong> texts is highly<br />
unlikely.<br />
Typographical errors are, however, a problem that must be taken<br />
seriously. Furthermore, errors <strong>of</strong>ten occur only once and consequently add to the<br />
number <strong>of</strong> hapaxes, and may therefore considerably affect the results (Lüdeling et<br />
al. 2000: 59). Plag expresses the same kind <strong>of</strong> concern, adding that some<br />
qualitative decisions are necessary even in purely quantitative analysis (1999: 29).<br />
Thus, in order to properly analyse the data for the present study, a certain amount<br />
<strong>of</strong> manual sorting was necessary.<br />
In general, corpora constitute a better source for the purposes <strong>of</strong> the<br />
present study. However, as has been shown, corpora as a source are far from<br />
unproblematic. Even the largest, most balanced and representative corpus will<br />
never equal real language use, even though it may provide a useful tool to make<br />
44
certain estimations or predictions based on statistics. As Baroni and Evert have<br />
put it, “corpora are <strong>fi</strong>nite samples from the in<strong>fi</strong>nite sets that constitutes a<br />
language” (2008: 777). <strong>The</strong>se samples let us make generalizations and inferences,<br />
but this requires that the problem at hand is accurately operationalized in<br />
quantitative terms (Evert and Baroni 2008: 777–778). A good strategy in the study<br />
<strong>of</strong> <strong>productivity</strong> is to combine corpus and dictionary evidence to provide a<br />
comprehensive view <strong>of</strong> the <strong>productivity</strong> <strong>of</strong> different patterns (see Plag 2006).<br />
3.3. How to determine units <strong>of</strong> analysis and other methodological problems<br />
One has to be careful when making inferences based on corpus evidence.<br />
Lüdeling et al. claim that even seemingly error-free corpora may cause trouble in<br />
this respect: for example, words may accidentally contain strings <strong>of</strong> characters<br />
that look like af<strong>fi</strong>xes but are not (2000: 59). For example, not all words that begin<br />
with sub- actually contain the pre<strong>fi</strong>x sub- in contemporary English (e.g., subtle),<br />
even if they might be derived from a word that was <strong>morphological</strong>ly complex in<br />
the donor language (Latin sub ‘under’ + tēla ‘web’ ) (Plag 2006: 543). In<br />
psycholinguistics, this phenomenon is referred to as pseudo-af<strong>fi</strong>xation (see, e.g.,<br />
Baayen 1993: 199). Another good example <strong>of</strong> pseudo-af<strong>fi</strong>xation is the status <strong>of</strong><br />
the constituent-fer in such words as infer, confer, prefer and transfer – it might be<br />
analysed as a bound root, but it does not carry a meaning that would be the same<br />
in all these words (see Plag 2003: 24–25). In the context <strong>of</strong> the present study,<br />
examples <strong>of</strong> words that seem to consist <strong>of</strong> several elements but that actually are<br />
mono-morphemic include hyperion and hyperoid. <strong>The</strong>refore, they are excluded<br />
from the type count.<br />
Another similar issue is the fact that many af<strong>fi</strong>xes have adopted<br />
idiosyncrasies that may blur their <strong>morphological</strong> make-up: for example, words<br />
that end in -ity occur in many transparent, complex words, but also in words like<br />
entity, which might not be considered complex (see also Plag 1999: 28). In a<br />
generous count, this word would probably be counted as a unit <strong>of</strong> analysis and<br />
included in the type count, but in that case it would skew the <strong>productivity</strong> rate <strong>of</strong> a<br />
modern word-formation process.<br />
45
Fortunately, the combining forms studied in this paper are quite<br />
distinctive in form, at least compared to af<strong>fi</strong>xes in the more traditional sense.<br />
However, several qualitative decisions have to be made concerning which<br />
elements to include and which to exclude from the type count. For example, Prćić<br />
points out that -logy and -holic can be analysed as combining forms in such words<br />
as morphology and foodaholic, but not in apology and alcoholic (2007: 382). In a<br />
similar vein, Warren suggests that hyper- can be considered a pre<strong>fi</strong>x in<br />
hyperactive but a combining form in hypertrophy (1990: 113). 20<br />
However, since it is not likely that speakers are aware <strong>of</strong> such a<br />
distinction, formations such as hypertrophy are not excluded from the data.<br />
Baayen points out that if an af<strong>fi</strong>x has several different, although related, senses,<br />
the goodness-<strong>of</strong>-<strong>fi</strong>t <strong>of</strong> the word frequency distribution might be lacking (2001:<br />
145). How the decision not to exclude the “pure neoclassical compound”<br />
instances <strong>of</strong> the <strong>selected</strong> combining forms affects their <strong>productivity</strong> values and the<br />
word frequency distributions will be assessed in the Discussion section.<br />
One <strong>of</strong> the problems <strong>of</strong> Baayen’s potential <strong>productivity</strong> index P is<br />
that none <strong>of</strong> the individual studies that apply the measure explains how the units<br />
<strong>of</strong> analysis are determined and how the data is treated in the study in question –<br />
this makes it dif<strong>fi</strong>cult to compare the <strong>productivity</strong> rates between different authors,<br />
even though they used the same data (see Plag 1999: 29). This is even more<br />
problematic with combining forms (at least if they are to be considered as a wordformation<br />
process distinct from af<strong>fi</strong>xation), since methods associated with<br />
af<strong>fi</strong>xation might not be fully applied to combining forms.<br />
In his study on the <strong>productivity</strong> <strong>of</strong> N + N compounds, Fernández-<br />
Domínguez deals with problems in applying the af<strong>fi</strong>x-centred <strong>productivity</strong><br />
measures to compounds in English (2009: 142). <strong>The</strong> problem with compounds is<br />
to determine which part corresponds to the base and which to the af<strong>fi</strong>x. <strong>The</strong><br />
problem can be tackled by for example calculating the average frequency <strong>of</strong> the<br />
components (Férnandez-Domínguez 2009: 143). However, in this study I will rely<br />
on the af<strong>fi</strong>x-based approach, since native compounds differ from neoclassical<br />
compounds and since combining forms are used in a more af<strong>fi</strong>x-like way (i.e.,<br />
they can be combined with free roots as well).<br />
20 In fact,Quirk et al. use the term combining form only with neoclassical compounds, and reserve<br />
the term neoclassical af<strong>fi</strong>x for the af<strong>fi</strong>x-like use <strong>of</strong> the same elements (1985: 1545).<br />
46
Determining the appropriate units <strong>of</strong> analysis is an important issue<br />
as well. For example, whether one counts ultra-violet and ultraviolet as a single<br />
type or separate types has consequences in terms <strong>of</strong> the P values one obtains.<br />
Sometimes an af<strong>fi</strong>x can also have several distinct, although to some extent<br />
interrelated meanings (see, e.g., McConchie 1998 for the different senses <strong>of</strong> the<br />
pre<strong>fi</strong>x dis- in English). Sometimes words generated with an apparently single af<strong>fi</strong>x<br />
are in fact generated by several distinct word-formation processes (Lüdeling and<br />
Evert 2003). For example, the German -lich may attach to adjectives, verbs and<br />
nouns alike. Plag also points out that lexicalization, i.e., a process in which a<br />
complex word develops a new, idiosyncratic meaning, may cause problems <strong>of</strong><br />
classi<strong>fi</strong>cation (2006: 543). However, the combining forms <strong>selected</strong> for this study<br />
turned out to be fairly unproblematic in this respect.<br />
In this study, I decided to count the spelling variants <strong>of</strong> a single<br />
lexeme (quasi-judicial vs. quasijudicial) as tokens <strong>of</strong> a single type. <strong>The</strong> spelling<br />
<strong>of</strong> English compounds is by and large unregulated, so that some compounds can<br />
have as many as three different spelling variants that Férnandez-Domínguez calls<br />
open (pseudo scienti<strong>fi</strong>c), solid (pseudoscienti<strong>fi</strong>c), and hyphenated (pseudoscienti<strong>fi</strong>c)<br />
(2009: 32–33). In a similar vein, words that have different spellings<br />
according to the different national (mainly British vs. American) spelling<br />
standards (e.g., palaeontology vs. paleontology) are counted as a single type.<br />
Of course, the number <strong>of</strong> types is not as relevant as the number <strong>of</strong><br />
tokens in this particular measure <strong>of</strong> <strong>productivity</strong>. However, the type hypertension,<br />
for example, occurs only once in the BNC, while hypertension has 369<br />
occurrences. Thus hyper-tension increases the <strong>productivity</strong> <strong>of</strong> hyper-, unless it be<br />
counted as a variant <strong>of</strong> hypertension. For a more detailed discussion <strong>of</strong> counting<br />
tokens and types, see Baroni (2008: 804–805).<br />
Multiply af<strong>fi</strong>xed words constitute another problem in determining<br />
the units <strong>of</strong> analysis. Should such word as conventionalizable be understood as an<br />
example <strong>of</strong> -ize or -able – or perhaps both (Plag 1999: 28–29)? <strong>The</strong> decision one<br />
makes has consequences for the <strong>productivity</strong> values and word frequency<br />
distributions <strong>of</strong> the elements <strong>selected</strong>. In the context <strong>of</strong> the present study, the<br />
problem is more relevant for <strong>fi</strong>nal than for initial combining forms.<br />
In the present study, I decided to count formations with a pre<strong>fi</strong>x or<br />
an additional initial combining form attached as separate types. <strong>The</strong>refore, neuro-<br />
47
physiology (with its spelling variants) and physiology are separate types, as well<br />
as aero-technology and technology. Another way would have been to add up the<br />
frequencies <strong>of</strong> aero-technology and technology. However, Baayen states that such<br />
cumulated token frequencies are more appropriate in estimating the activation<br />
level A than in calculating the values for P or P* for at least two reasons. First, it<br />
is dif<strong>fi</strong>cult to estimate the weight <strong>of</strong> the various word-formation processes in<br />
multiply complex words. Second, if a word is assumed to be an independent free<br />
form in the lexicon, it will be counterproductive to sum up the token frequencies<br />
<strong>of</strong> these two (1993: 200–201). If a word were assigned to two <strong>morphological</strong><br />
categories, the observations in those two categories would no longer be<br />
independent, and the statistical tests for comparing <strong>productivity</strong> rates would no<br />
longer be valid (Baayen 2008: 903). In their study on the <strong>productivity</strong> <strong>of</strong> <strong>selected</strong><br />
nominal suf<strong>fi</strong>xes (such as -al and -ist), Plag et al. have chosen to exclude pre<strong>fi</strong>xed<br />
formations (unavailable) and compounds (performance-artist) completely (1999:<br />
214).<br />
It must be noted again that in some respect the rules concerning<br />
af<strong>fi</strong>xes cannot be fully applied to the study <strong>of</strong> combining forms. For example, it is<br />
not always easy to determine which component is the base and which one<br />
corresponds to the af<strong>fi</strong>x. However, it is intuitively clear that adding a pre<strong>fi</strong>x (aero-<br />
) to a neoclassical compound (technology) involves lexical creativity, and thus<br />
contributes to the <strong>productivity</strong> <strong>of</strong> -logy. On the other hand, since formations with<br />
neoclassical compounds are unclear cases in many ways, the opposite view could<br />
be justi<strong>fi</strong>ed as well.<br />
It must also be noted that there are several <strong>fi</strong>nal combining forms<br />
that have related forms, such as -logical from -logy and -graphical or -graphic<br />
from -graphy (called suf<strong>fi</strong>xal expansion by Prćić (2007: 386–387)). <strong>The</strong>se forms<br />
will not be dealt with in this study, for the reasons mentioned above. P is a<br />
category-conditioned degree <strong>of</strong> <strong>productivity</strong> and is thus calculated on the basis <strong>of</strong><br />
a single <strong>morphological</strong> category (see Baayen 1993: 200).<br />
48
4. Results<br />
4.1. Potential <strong>productivity</strong> P<br />
In order to <strong>fi</strong>nd all the instances <strong>of</strong> the initial combining forms, the wildcard * was<br />
used to include all the formations that begin with a certain sequence <strong>of</strong> letters, in<br />
this case hyper*, ultra*, pseudo*, and quasi*. With respect to <strong>fi</strong>nal combining<br />
forms, the query was formed as *log[y,ies], *graph[y,ies], etc., in order to include<br />
the plural form <strong>of</strong> the <strong>selected</strong> combining forms as well, which were all nounforming.<br />
<strong>The</strong> queries were restricted to cover only the written part <strong>of</strong> the BNC.<br />
This makes it easier to compare the results to those obtained in previous studies<br />
(e.g., Plag 1999, 2002, 2006).<br />
Table 2 summarizes the type frequency V, the token frequency N, as<br />
well as the number <strong>of</strong> hapax legomena n 1 and the values <strong>of</strong> potential <strong>productivity</strong><br />
P <strong>of</strong> each combining form that was <strong>selected</strong> for this study.<br />
Table 2. Type frequency V, token frequency N, the number <strong>of</strong> hapax legomena n 1,<br />
and the <strong>productivity</strong> rate P <strong>of</strong> <strong>selected</strong> combining forms in the written part <strong>of</strong> the<br />
BNC.<br />
V N n 1 P<br />
hyper- 323 2,859 163 0.0570<br />
ultra- 317 1,628 188 0.1155<br />
pseudo-<br />
325<br />
818<br />
231<br />
0.2824<br />
-logy<br />
375<br />
36,077<br />
111<br />
0.0031<br />
-graphy<br />
166<br />
7,440<br />
57<br />
0.0077<br />
-nomy 46 14,240 20 0.0014<br />
In terms <strong>of</strong> type frequency V, the initial combining forms generally<br />
seem to rank higher than the <strong>fi</strong>nal combining forms. <strong>The</strong> only exception is -logy,<br />
which has the highest type frequency <strong>of</strong> all the <strong>selected</strong> combining forms. On the<br />
49
other hand, the <strong>fi</strong>nal combining form -nomy has a particularly low type frequency,<br />
when compared to the others. It must be noted that the difference in type<br />
frequencies between initial and <strong>fi</strong>nal combining forms would be even more drastic<br />
if multiply-pre<strong>fi</strong>xed words were counted as a single type. In this study, such<br />
words as aero-technology and gastro-technology (both hapaxes) have been<br />
counted as separate types, not as occurrences <strong>of</strong> technology, and thus the type<br />
frequency (and consequently the number <strong>of</strong> hapax legomena) is higher than with a<br />
less generous count.<br />
Some tentative conclusions can be already made by looking at the<br />
type frequencies <strong>of</strong> the <strong>selected</strong> combining forms, since they clearly measure<br />
some aspect <strong>of</strong> <strong>productivity</strong>. As has been stated earlier, this aspect has been called<br />
pr<strong>of</strong>itability, extent <strong>of</strong> use, or realized <strong>productivity</strong>. Thus, at this point, one could<br />
suggest that -logy has a high extent <strong>of</strong> use, while with -nomy the extent is not so<br />
substantial. <strong>The</strong> three initial combining forms and -graphy are situated somewhere<br />
in between on the scale.<br />
As regards the token frequency N, there is more variation, and again<br />
some interesting tendencies can be observed there as well, even though token<br />
frequency in itself has not been generally considered as a very good direct<br />
measure <strong>of</strong> <strong>productivity</strong> (see, e.g., Bauer 2001: 147). <strong>The</strong> most frequent combining<br />
form is -logy, with 36,077 occurrences, which is more than twice as frequent as the<br />
next frequent combining form, which is -nomy with 14,240 occurrences. Both<br />
have some extremely frequent types that contribute greatly to their total token<br />
frequencies. For example, technology has 11,353 occurrences, which equals more<br />
than 30 percent <strong>of</strong> the total number <strong>of</strong> tokens. If formations with technology plus a<br />
pre<strong>fi</strong>xal element or another initial combining form attached (such as aerotechnology<br />
or multi-technology, both hapaxes) are counted as a single type, the<br />
total token frequency is even higher. As regards -nomy, the relative weight <strong>of</strong> the<br />
most frequent type (economy with 9,997 occurrences) is even more substantial<br />
(approximately 70 percent <strong>of</strong> all the tokens with -nomy). In general, the initial<br />
combining forms have far lower token frequencies than the <strong>fi</strong>nal combining<br />
forms.<br />
In terms <strong>of</strong> the number <strong>of</strong> hapax legomena n 1, the initial combining<br />
forms display much higher frequencies than the <strong>fi</strong>nal combining forms. <strong>The</strong><br />
hapaxes also include many lexemes that are not attested in the OED. <strong>The</strong>se<br />
50
include such items as hyperalertness, pseudo-Scottish, and ultra-Zionist. As for<br />
the <strong>fi</strong>nal combining forms, such hapaxes include ghettology and dreamography.<br />
<strong>The</strong>re are at least two reasons why they are not listed in the dictionary. Firstly,<br />
they are quite rare in the language, and secondly, they are semantically rather<br />
transparent. For example, hyperalertness need not be listed in a dictionary,<br />
because alertness already has an entry, and because its meaning can be constituted<br />
from the meaning <strong>of</strong> its parts. This can be seen as another indicator for the<br />
<strong>productivity</strong> <strong>of</strong> the combining forms examined, since <strong>productivity</strong> and<br />
transparency are at least indirectly linked (see Dalton-Puffer 1994: 252–253,<br />
Pounder 2000: 134). As was stated earlier, transparent, infrequent words are<br />
parsed, instead <strong>of</strong> being accessed as a whole (Frauenfelder and Schreuder 1992).<br />
<strong>The</strong> only exception in this group <strong>of</strong> words is -nomy, since all hapaxes with -nomy<br />
have an entry in the OED. Besides, most <strong>of</strong> them are scienti<strong>fi</strong>c words, such as<br />
chirognomy and aeronomy.<br />
At this point, it might be justi<strong>fi</strong>ed to examine the ratio <strong>of</strong> the number<br />
<strong>of</strong> hapaxes to the overall type frequency <strong>of</strong> each combining form. This ratio has<br />
not been used in any <strong>of</strong> the previous literature as a measure <strong>of</strong> <strong>morphological</strong><br />
<strong>productivity</strong>, but it might <strong>of</strong>fer some tentative insights on the relative importance<br />
<strong>of</strong> hapaxes. In this sense, it certainly describes some aspect <strong>of</strong> <strong>productivity</strong> by<br />
being indirectly linked to potential <strong>productivity</strong> P.<br />
Table 3 displays the proportion <strong>of</strong> hapax legomena out <strong>of</strong> the total<br />
type frequency for each combining form. <strong>The</strong> ratio is higher with initial<br />
combining forms, being over 50 per cent in all the cases. <strong>The</strong> <strong>fi</strong>nal combining<br />
forms do not seem to lag very far behind, however. <strong>The</strong> table can help make<br />
predictions about the P values <strong>of</strong> the combining forms, which will be discussed<br />
next.<br />
51
Table 3. <strong>The</strong> ratio <strong>of</strong> hapax legomena n 1 to the type frequency V <strong>of</strong> <strong>selected</strong><br />
combining forms in the written part <strong>of</strong> the BNC.<br />
n 1 / V<br />
hyper- 0.5046<br />
ultra- 0.5931<br />
pseudo-<br />
-logy<br />
-graphy<br />
0.7108<br />
0.296<br />
0.3434<br />
-nomy 0.4347<br />
<strong>The</strong> last column in Table 2 presents the P values for the <strong>selected</strong><br />
combining forms. Again, the initial combining forms score higher. <strong>The</strong> lower P<br />
values <strong>of</strong> the <strong>fi</strong>nal combining forms are partly explained by the presence <strong>of</strong> those<br />
few very frequent types mentioned above. With the initial combining forms, such<br />
extremely frequent types are nonexistent. For example, with hyper-, which is the<br />
least productive initial combining form as regards potential <strong>productivity</strong> P, the<br />
most frequent type is hypertension (368 occurrences) and with ultra-, ultraviolet<br />
(332 occurrences). <strong>The</strong>se are only a fraction <strong>of</strong> the corresponding numbers for the<br />
<strong>fi</strong>nal combining forms.<br />
What can be said about the overall <strong>productivity</strong> <strong>of</strong> the <strong>selected</strong><br />
combining forms? <strong>The</strong> problem with interpreting the results is that the P values<br />
alone cannot give us any de<strong>fi</strong>nitive answers. Baayen and Lieber emphasize the<br />
fact that P is only a relative measure and cannot therefore in itself be used to<br />
determine whether a particular word-formation rule is productive or not (1991:<br />
816). <strong>The</strong> method is statistical, but it does not give any threshold values that<br />
would indicate the boundaries between productive and unproductive processes. As<br />
Plag has noted, P only provides “a continuum <strong>of</strong> more or less productive<br />
processes”, so that it is dif<strong>fi</strong>cult or even impossible to pin down where exactly the<br />
productive processes end and the unproductive processes begin (1999: 33).<br />
Baayen and Lieber suggest that in order to decide whether a process<br />
is productive, its P value should be compared to the P value <strong>of</strong> a relevant set <strong>of</strong><br />
52
simplex words (1991: 816). However, calculating the P value <strong>of</strong>, say, an<br />
inflectional process would be beyond the scope <strong>of</strong> this study. Another way to<br />
estimate the <strong>productivity</strong> <strong>of</strong> the <strong>selected</strong> combining forms is to compare it to a set<br />
<strong>of</strong> words formed with another word-formation process, such as suf<strong>fi</strong>xes as<br />
opposed to pre<strong>fi</strong>xes, or vice versa. Plag, for example, provides the P values for<br />
seven different suf<strong>fi</strong>xes in English (2006: 545, based on data from Plag et al. 1999<br />
and Plag 2002). <strong>The</strong> data come from the written part <strong>of</strong> the BNC as well (see Plag<br />
et al. 1999: 210), so the <strong>fi</strong>gures should be more or less comparable to those <strong>of</strong> the<br />
present study, even though Plag does not <strong>of</strong>fer very rigorous details about the<br />
basis on which he has determined the units <strong>of</strong> analysis (see section 3).<br />
Table 4. Type frequency V, token frequency N, the number <strong>of</strong> hapax legomena n 1,<br />
and the <strong>productivity</strong> rate P <strong>of</strong> <strong>selected</strong> suf<strong>fi</strong>xes in the written part <strong>of</strong> the BNC.<br />
From Plag (2006: 545).<br />
V N n 1 P<br />
-ness 2,466 1,369,116 943 0.061<br />
-ion 2,392 371,747 524 0.0338<br />
-ity 1,372 106,957 354 0.0096<br />
-ist<br />
1,207<br />
98,823<br />
341<br />
0.0088<br />
-less<br />
682<br />
28,340<br />
272<br />
0.0036<br />
-ish<br />
491<br />
7,745<br />
262<br />
0.00092<br />
-wise 183 2,091 128 0.00038<br />
Table 4 presents the type and token frequencies as well as the<br />
number <strong>of</strong> hapax legomena and the potential <strong>productivity</strong> values <strong>of</strong> <strong>selected</strong><br />
suf<strong>fi</strong>xes in the written part <strong>of</strong> the BNC, calculated by Plag (2006: 545). <strong>The</strong> suf<strong>fi</strong>x<br />
-wise has the highest P value (0.061), and -ion the lowest (0.00038). If these<br />
values are compared to those <strong>of</strong> the combining forms that were examined in this<br />
study, it can be seen that the suf<strong>fi</strong>xes have signi<strong>fi</strong>cantly lower P values than the<br />
combining forms do. Even the least productive pre<strong>fi</strong>x, hyper-, (P = 0.0570) has a<br />
53
somewhat higher P value than the most productive suf<strong>fi</strong>x in Plag’s study. As for<br />
the <strong>fi</strong>nal combining forms, they score lower than the most productive suf<strong>fi</strong>xes, but<br />
higher than the least productive ones. One <strong>of</strong> the disadvantages with Plag’s study<br />
is that is since it does not account for pre<strong>fi</strong>xes, it is not possible to see whether<br />
pre<strong>fi</strong>xes in general would rank higher than suf<strong>fi</strong>xes.<br />
To summarize, the initial combining forms seem to rank somewhat<br />
higher than the <strong>fi</strong>nal combining forms, both in terms <strong>of</strong> the number <strong>of</strong> different<br />
types V and potential <strong>productivity</strong> P. On the other hand, the <strong>fi</strong>nal combining forms<br />
in general have higher token frequencies N, which is <strong>of</strong>ten caused by the existence<br />
<strong>of</strong> a few very frequent types. What this section has discussed constitutes the basic<br />
<strong>productivity</strong> measures that have been established and widely adopted in the study<br />
<strong>of</strong> <strong>productivity</strong>. <strong>The</strong>re is, however, a set <strong>of</strong> measures more rarely employed in<br />
studies on the <strong>productivity</strong> <strong>of</strong> different word-formation processes. <strong>The</strong>se measures<br />
rely on lexical statistics and word frequency distributions, and thus illustrate a<br />
slightly different aspect <strong>of</strong> <strong>productivity</strong>.<br />
4.2. LNRE modelling<br />
It was stated above that Zipf(-Mandelbrot)’s law <strong>fi</strong>ts the word frequency<br />
distribution <strong>of</strong> naturally occurring texts reasonably well. <strong>The</strong> same holds true for<br />
productive word-formation processes as well, which makes LNRE modelling<br />
particularly useful for the study <strong>of</strong> <strong>productivity</strong>. <strong>The</strong>re are several methods that<br />
can be used to obtain information on the <strong>productivity</strong> <strong>of</strong> the combining forms<br />
<strong>selected</strong>.<br />
As was stated above, one <strong>of</strong> the inherent problems with P is that the<br />
proportion <strong>of</strong> hapax legomena does not grow linearily as the corpus size N<br />
increases. This makes it dif<strong>fi</strong>cult to compare the P values between different wordformation<br />
processes, since the sample sizes for each process varies. A solution<br />
that was <strong>of</strong>fered in the previous section was to compare the sets <strong>of</strong> P values <strong>of</strong><br />
different word-formation processes. However, there is a more elegant way to<br />
determine the <strong>productivity</strong> <strong>of</strong> the combining forms: the zipfR package <strong>of</strong> the R<br />
statistics s<strong>of</strong>tware (Evert and Baroni 2007) allows the prediction <strong>of</strong> the shape <strong>of</strong><br />
54
the vocabulary growth curve at arbitrary sample sizes. 21 Another method that is<br />
particularly useful in the study <strong>of</strong> <strong>productivity</strong> is calculating the expected<br />
frequency spectra for the combining forms and comparing them to the observed<br />
values. This allows us to assess the goodness-<strong>of</strong>-<strong>fi</strong>t <strong>of</strong> the statistical model, i.e.,<br />
how well it <strong>fi</strong>ts the data. <strong>The</strong>refore, it can help us detect possible outliers, i.e., data<br />
points that deviate signi<strong>fi</strong>cantly from the rest <strong>of</strong> the sample.<br />
Table 5 presents the observed and expected P values for each<br />
combining form. For -logy, the values are identical, since -logy has the highest<br />
token frequency and therefore serves as the “base” against which the other<br />
combining forms are compared. For the other combining forms, the differences<br />
between the observed and expected values vary, depending on their sample size.<br />
In addition to their P values, the ordering <strong>of</strong> the combining forms after the<br />
extrapolation operation changes as well. However, pseudo- has the highest P<br />
value in both cases, and -nomy the lowest.<br />
Table 5. <strong>The</strong> observed and expected P values for the <strong>selected</strong> combining forms in<br />
the written part <strong>of</strong> the BNC.<br />
P<br />
P (expected)<br />
(observed)<br />
hyper- 0.0570 0.0052<br />
ultra- 0.1155 0.0047<br />
pseudo-<br />
-logy<br />
-graphy<br />
0.2824<br />
0.0031<br />
0.0077<br />
0.0347<br />
0.0031<br />
0.0028<br />
-nomy 0.0014 0.0007<br />
21 For a sample session illustrating the necessary operations for obtaining the expected P values for<br />
different word-formation processes, see Evert and Baroni 2007.<br />
55
A more visual way to compare the <strong>productivity</strong> rates is to look at<br />
their vocabulary growth curves, which are shown in <strong>fi</strong>gure 4. A vocabulary<br />
growth curve can be used to show changes in vocabulary size V as the sample size<br />
N increases (Evert and Baroni 2007: 32). <strong>The</strong> steeper the curve, the more rapidly<br />
the vocabulary (i.e., the number <strong>of</strong> types) is growing, and the more productive a<br />
particular word-formation process can be considered. In a sense, a vocabulary<br />
growth curve works in a similar way to the potential <strong>productivity</strong> P, since both<br />
attempt to capture the rate at which the vocabulary <strong>of</strong> a given word-formation<br />
process grows. <strong>The</strong> only difference is that a vocabulary growth curve relies on<br />
types, while P makes use <strong>of</strong> hapax legomena.<br />
As can be seen in Figure 4, pseudo- holds the highest place, which is<br />
shown by the steepest curve. On the other hand, -nomy has the lowest<br />
<strong>productivity</strong>, which is represented by a moderate, almost flat curve. In general, the<br />
initial combining forms retain the highest <strong>productivity</strong> ranks, while all the <strong>fi</strong>nal<br />
combining forms cluster at the low end <strong>of</strong> the graph. All the slopes follow the<br />
general tendency to grow more rapidly at low N and then gradually decrease. At<br />
small sample sizes, it is more likely to encounter a yet unattested type, but the<br />
likelihood decreases as the sample size increases.<br />
56
Figure 4. Extrapolated vocabulary growth curves for the <strong>selected</strong> combining<br />
forms. <strong>The</strong> x axis displays corpus size N, and the y axis the expected type<br />
frequencies at different sample sizes.<br />
57
Figure 5. <strong>The</strong> expected (black) versus observed (red) frequency spectra for the <strong>selected</strong><br />
combining forms. <strong>The</strong> x axis displays frequency classes, and the y axis displays the<br />
number <strong>of</strong> types in each class.<br />
58
Not only vocabulary growth curves, but also frequency spectra<br />
provide a useful tool to assess the <strong>productivity</strong> <strong>of</strong> different word-formation<br />
processes. In particular, a frequency spectrum provides information on the<br />
distribution <strong>of</strong> the type frequencies <strong>of</strong> a given word-formation process, because<br />
productive ones tend to behave in a certain way in this respect, producing a<br />
skewed, rapidly decreasing plot. Even looking at the frequency spectra per se can<br />
reveal interesting facts, but it is even more useful to compare the observed values<br />
to the expected, average values. Again, this can be done with the ZipfR package<br />
<strong>of</strong> the R statistics s<strong>of</strong>tware. <strong>The</strong> actual formula to calculate the expected values for<br />
a frequency spectrum is explained by Baayen (2001: 92).<br />
Figure 5 shows the expected (black) versus observed (red) frequency<br />
spectra for the combining form data. <strong>The</strong> <strong>fi</strong>gure only displays the <strong>fi</strong>rst <strong>fi</strong>fteen<br />
frequency classes, but it can nevertheless give a useful visualization <strong>of</strong> how the<br />
different types are distributed with each combining form. Even a brief glance at<br />
the <strong>fi</strong>gure shows that all the combining forms display similar, rapidly decreasing<br />
plots with a long tail <strong>of</strong> high frequency classes that are realized by only few types.<br />
In addition, as the <strong>fi</strong>gure shows, almost all the observed frequency spectra line up<br />
relatively beautifully with the expected, average values.<br />
<strong>The</strong> only exception is -nomy, which displays quite drastic<br />
differences between the expected and the observed values at times. For example,<br />
frequency class 2 (i.e., the number <strong>of</strong> types that occur twice in the corpus) has<br />
many more members than expected. <strong>The</strong> same holds true for frequency class 5.<br />
Scarcity <strong>of</strong> data cannot be the reason for this inconsistency, since the token<br />
frequency <strong>of</strong> -nomy is the second largest <strong>of</strong> all the <strong>selected</strong> combining forms. On<br />
the other hand, despite its high token frequency, -nomy has a particularly low type<br />
frequency, only 46 types (the most frequent combining forms having over 300<br />
types). In a situation like this, even a small deviation from the expected values<br />
might have an impact. Another striking feature with -nomy, which might have an<br />
impact on the results is that the most frequent type, economy, accounts for over 70<br />
per cent <strong>of</strong> all the words with -nomy. With -graphy, the corresponding percentage<br />
is roughly 20% (the most frequent type being geography), and with hyper, no<br />
more than 13% (hypertension).<br />
59
One explanation for the lack <strong>of</strong> goodness-<strong>of</strong>-<strong>fi</strong>t might also be<br />
sampling error, in which case the irregularities would be outliers, i.e., atypical<br />
data points with exceptionally high or low values (see Baayen 2008: 91).<br />
However, one has to be careful when discarding data as potential outliers. As<br />
Baayen has put it, “you should not tweak the data by removing data points so that<br />
a non-signi<strong>fi</strong>cant effect becomes signi<strong>fi</strong>cant” (2008: 237). Baayen also points out<br />
that the statistical LNRE models rely on the assumption that words occur<br />
randomly and independently <strong>of</strong> each other in texts, which is <strong>of</strong> course a<br />
simpli<strong>fi</strong>cation and might cause the lack <strong>of</strong> goodness-<strong>of</strong>-<strong>fi</strong>t (2008: 233). Of course,<br />
the randomness factor might not be as pervasive in lexical studies as in studies on<br />
syntax.<br />
<strong>The</strong> slight out-<strong>of</strong>-datedness <strong>of</strong> the BNC mentioned in section 3.2 is a<br />
factor that must not be overlooked either. In fact, there are several cases where a<br />
more up-to-date corpus would certainly have yielded different results. <strong>The</strong> word<br />
euro-cracy <strong>of</strong>fers a good example, even though it is not among the combining<br />
forms <strong>selected</strong> for this study. It is a hapax in the BNC, but in view <strong>of</strong> the extralinguistic<br />
development <strong>of</strong> the last two decades, it is certainly not a very rare word<br />
in British English. On the other hand, eurocracy also turned out to be a hapax in<br />
the COCA corpus, which is updated every year and should thus cover even the<br />
most recent changes in the English language. However, this might partially be<br />
explained by the fact that the COCA records American English. <strong>The</strong>refore the out<strong>of</strong>-date<br />
nature <strong>of</strong> the BNC shows up and skews the results. It is also possible to<br />
<strong>fi</strong>nd examples <strong>of</strong> words which have been fashionable at the moment <strong>of</strong> corpus<br />
compilation, but which have since then become rare or even obsolete, as a result<br />
<strong>of</strong> changes in fashion or technological development. Hypersparc (100<br />
occurrences), which refers to a 1990’s microprocessor, is a good example.<br />
5. Discussion<br />
It is a relatively easy task to extract <strong>fi</strong>gures and tables from the data and count the<br />
P values for a set <strong>of</strong> word-formation processes – even if the methodological<br />
problems discussed in section 3.3 are taken into consideration. In a similar vein,<br />
60
calculating frequency spectra and vocabulary growth curves with R is a relatively<br />
straightforward process. In fact, in many studies that rely on P as an indicator <strong>of</strong><br />
<strong>productivity</strong>, the discussion <strong>of</strong> results is rather super<strong>fi</strong>cial and the evidence drawn<br />
from the P values <strong>of</strong> af<strong>fi</strong>xes is taken for granted.<br />
However, careful qualitative analysis <strong>of</strong> the results is necessary in<br />
order to get a realistic view <strong>of</strong> the <strong>productivity</strong> <strong>of</strong> these combining forms.<br />
According to Evert and Lüdeling, quantitative and qualitative analyses can<br />
complement each other, and while quantitative approaches provide useful<br />
information on the nature <strong>of</strong> different word-formation processes, they cannot<br />
really be used without qualitative interpretation (2000: 167–168). <strong>The</strong>re are<br />
several linguistic and extra-linguistic factors that must be taken into account. <strong>The</strong><br />
factors may also shed light on eventual differences in <strong>productivity</strong> rates obtained<br />
with various <strong>productivity</strong> measures. It is also worthwhile to discuss the reasons<br />
that might explain those differences. Issues like these will be the topic <strong>of</strong> this<br />
section.<br />
In general, all the measures employed in this study seem to indicate<br />
that the initial combining forms have higher <strong>productivity</strong> rates than the <strong>fi</strong>nal<br />
combining forms. <strong>The</strong>y have higher P values, and they produce steeper<br />
vocabulary growth curves. On the other hand, it seems that there are no major<br />
differences in the goodness-<strong>of</strong>-<strong>fi</strong>t <strong>of</strong> the frequency spectra between the initial and<br />
the <strong>fi</strong>nal combining forms comparing the observed and expected frequency<br />
spectra. However, based on the evidence brought about by potential <strong>productivity</strong><br />
P and vocabulary growth curves, we might well claim that that the initial<br />
combining forms score higher than the <strong>fi</strong>nal combining forms.<br />
What might the reasons for the higher <strong>productivity</strong> rates obtained by<br />
the initial combining forms be? <strong>The</strong> differences between initial and <strong>fi</strong>nal<br />
combining forms from a semantic or structural point <strong>of</strong> view have not been<br />
considered in previous literature, but the pre<strong>fi</strong>x-suf<strong>fi</strong>x distinction might prove a<br />
useful analogy, since it is not certain that speakers make a distinction between<br />
af<strong>fi</strong>xes and combining forms. Nevalainen writes that pre<strong>fi</strong>xes have fewer wordclass<br />
restrictions on their input range than suf<strong>fi</strong>xes do (1999: 355). From the<br />
semantic point <strong>of</strong> view, the main linguistic function <strong>of</strong> pre<strong>fi</strong>xes is <strong>of</strong>ten semantic,<br />
while suf<strong>fi</strong>xes tend to be class-changing, but more abstract in meaning (see<br />
section 2.1.).<br />
61
Studies concentrating on or touching upon the <strong>productivity</strong> <strong>of</strong><br />
combining forms have mainly been theoretical. Warren, for example, rather<br />
cryptically writes that pre<strong>fi</strong>xes have “productive force”, while combining forms<br />
“need not have productive force” but can be nonce formations (1990: 123). 22<br />
Similarly, Prćić states that the <strong>productivity</strong> <strong>of</strong> pre<strong>fi</strong>xes is systematic, since they are<br />
used regularly in “ready-made morpho-syntactic patterns”, and that each pre<strong>fi</strong>x<br />
can be assigned a value <strong>of</strong> high, restricted, or low <strong>productivity</strong> (2005: 325).<br />
Combining forms, on the other hand, “are simply there to be used if/when need<br />
for them arises”, and thus display non-systematic <strong>productivity</strong>, comparable to that<br />
<strong>of</strong> elements in compounding (Prćić 2005: 325). Fischer states that dictionaries can<br />
be used to assess the <strong>productivity</strong> <strong>of</strong> combining forms since if a combining form<br />
has its own entry in a dictionary, it can be said to have at least some degree <strong>of</strong><br />
<strong>productivity</strong> (1998: 63). However, the fact that all the elements studied in this<br />
paper have an entry in the OED is too weak an indicator in itself to gauge their<br />
<strong>productivity</strong>.<br />
It is also worthwhile to consider the general tendencies or<br />
underlying changes that might be going on in terms <strong>of</strong> the English word-formation<br />
system. Bauer argues that since many <strong>of</strong> the native pre<strong>fi</strong>xes in English have lost<br />
ground, having been replaced by learned “pre<strong>fi</strong>x-like elements”, a gradual<br />
typological shift from pre<strong>fi</strong>xation towards “something more like compounding”<br />
might be taking place (2003a: 35, 37). Even though he does not state it explicitly,<br />
by these “learned pre<strong>fi</strong>x-like elements” Bauer seems to speci<strong>fi</strong>cally mean<br />
neoclassical combining forms. Szymanek seems to hold a similar view as well,<br />
claiming that the neoclassical compounds have gained ground in the English<br />
word-formation system, especially during the last few decades (2005: 432).<br />
However, Bauer and Szymanek do not discuss whether these tendencies apply to<br />
suf<strong>fi</strong>xes and <strong>fi</strong>nal combining forms as well.<br />
Szymanek also points out that coining new words, at least nouns, by<br />
compounding, is easy for speakers, due to recursion and the absence <strong>of</strong> any major<br />
grammatical restrictions (2005: 432). <strong>The</strong> same holds true for pre<strong>fi</strong>xes, as they,<br />
22 Bauer (1983: 45–46) de<strong>fi</strong>nes nonce formations as complex words that are created “on the spur <strong>of</strong><br />
the moment”, usually to cover an immediate need. Thus they are usually unique and rare. Nonce<br />
formations must not be confused with hapax legomena: hapaxes are always de<strong>fi</strong>ned in terms <strong>of</strong> a<br />
certain corpus. When nonce formations start to be used regularly in the speech community and<br />
become institutionalized, they become neologisms (Fischer 1998: 15–16).<br />
62
too, can be used recursively, and usually do not affect the stress pattern <strong>of</strong> their<br />
bases and do not have segmental phonological effects on their bases (Lieber 2005:<br />
389).<br />
In a similar vein, as mentioned earlier, Hughes has noted that many<br />
learned pre<strong>fi</strong>xes, such as extra-, super- and hyper- have gained new, creative<br />
meanings and are thus used with greater flexibility in contemporary English than<br />
they were before (2003: 346–347). Yet another fact suggesting the high<br />
<strong>productivity</strong> <strong>of</strong> pre<strong>fi</strong>xes or initial combining forms comes from Bauer, who uses<br />
the term word families to illustrate the fact that a single pre<strong>fi</strong>xing operation may<br />
result in several members <strong>of</strong> the same family (as in biodegradability,<br />
biodegradable, biodegrade, biodegradation), which <strong>of</strong> course increases the P<br />
value <strong>of</strong> the element in question. Kastovsky notes that besides scienti<strong>fi</strong>c<br />
terminology, neologisms that exploit combining forms also occur in science<br />
<strong>fi</strong>ction novels, which also demonstrates the <strong>productivity</strong> <strong>of</strong> this word-formation<br />
process (2009: 1). He mentions such examples as cryosleep, holomovie, and exobiology.<br />
With these observations in mind, combined with evidence from the BNC,<br />
it is not too bold to state that all the combining forms studied in this paper are<br />
fairly productive.<br />
Etymology is a factor that might play an important role in the<br />
differences between the <strong>productivity</strong> rates between initial and <strong>fi</strong>nal combining<br />
forms. <strong>The</strong> latter all come from lexical words (mostly nouns), while some initial<br />
combining forms come from pre<strong>fi</strong>xes (hyper) – do they resemble pre<strong>fi</strong>xes in other<br />
senses as well? Another question is how readily the <strong>selected</strong> combining forms<br />
combine with native bases as opposed to Greek and Latin elements – a possible<br />
topic for a future study.<br />
All this raises the question whether speakers <strong>of</strong> English actually<br />
differentiate between af<strong>fi</strong>xes and combining forms in general and whether the<br />
distinction between these two word-formation processes has any relevance from<br />
the point <strong>of</strong> view <strong>of</strong> cognitive processing and the mental lexicon. As has already<br />
been demonstrated above, it is not always easy to distinguish between af<strong>fi</strong>xes and<br />
combining forms (or combining forms and bases) even in terms <strong>of</strong> <strong>morphological</strong><br />
theory. <strong>The</strong> distinction between pre<strong>fi</strong>xes and initial combining forms in particular<br />
does not seem to be so clear-cut. Prćić states that modern morphology in general<br />
has been unable to make a consistent distinction between these two types <strong>of</strong> word-<br />
63
formation, and they are also <strong>of</strong>ten labelled inconsistently in dictionaries as well<br />
(2005: 313–314).<br />
Marchand writes that learned af<strong>fi</strong>xes and combining forms are “on<br />
the same footing” in the English word-formation system, since they are both<br />
borrowed elements that do not exist independently as words (1969: 131–132).<br />
Similarly, Bauer, when suggesting that English is in the process <strong>of</strong> going through<br />
a gradual typological change, in fact de<strong>fi</strong>nes pre<strong>fi</strong>xes and initial combining forms<br />
as a single category <strong>of</strong> “pre<strong>fi</strong>xal elements” and treats elements like anti-, sub-,<br />
super-, mini- and micro- as members <strong>of</strong> a similar category, although the OED<br />
de<strong>fi</strong>nes the <strong>fi</strong>rst three as pre<strong>fi</strong>xes, and the latter two as combining forms (2003a).<br />
In a similar manner, Nevalainen seems to treat all the pre<strong>fi</strong>xal elements, native<br />
and non-native alike, as a single category that she calls pre<strong>fi</strong>xes (1999: 383).<br />
Prćić suggests that certain elements that have traditionally been<br />
de<strong>fi</strong>ned as combining forms (including pseudo- and quasi-) may become more<br />
like af<strong>fi</strong>xes after repeated use: they acquire a “ready-made morphosemantic<br />
pattern” (2005: 329). 23 He goes as far as to suggest that elements like mega-,<br />
micro-, multi-, pseudo- and quasi-should be recategorized as pre<strong>fi</strong>xes for various<br />
reasons, one <strong>of</strong> them being the fact that they modify the meaning <strong>of</strong> the base in a<br />
regular way (2005: 329). Warren has come to a similar conclusion, claiming that<br />
af<strong>fi</strong>xes traditionally result from a grammaticalization process while combining<br />
forms were originally bound lexical elements und thus “better equipped to resist<br />
grammaticalization” (1990: 123–124). However, she also writes that combining<br />
forms are not immune to grammaticalization either, and thus some <strong>of</strong> them have<br />
moved towards an af<strong>fi</strong>xal status. Fundamentally, this all comes down to<br />
Kastovsky’s claim (2009) that that a category like combining forms is altogether<br />
unnecessary and that processes like af<strong>fi</strong>xation, compounding, blending and<br />
clipping could be used to deal with that category.<br />
In any case, it is evident that <strong>productivity</strong> in the strict sense only<br />
measures one aspect <strong>of</strong> <strong>productivity</strong>. A more comprehensive study would involve<br />
a combination <strong>of</strong> different methods. Baayen and Lieber themselves acknowledge<br />
the fact that the type frequency <strong>of</strong> a given af<strong>fi</strong>x is also an indicator <strong>of</strong> its extent <strong>of</strong><br />
23 Carstairs-McCarthy writes that the recent proliferation <strong>of</strong> combining forms in association with<br />
free Germanic roots can actually be seen as a rather peculiar trend, since Latin- and Greek-derived<br />
words are otherwise quite rare in present-day English (2002: 109).<br />
64
use and may thus be <strong>of</strong> interest in such <strong>fi</strong>elds as lexicography or applied<br />
linguistics (1991: 817).<br />
6. Conclusion<br />
Sometimes ignorant but pretentious people take to<br />
coining words, re-interpreting foreign words in their own<br />
way. <strong>The</strong>y vaguely feel that there is some characteristic<br />
termination in a Greek or Latin word which they then<br />
attach to some English basis to give a combination a<br />
‘learned’ tinge. As a result, we get barbarisms in -athon,<br />
coined after Marathon, such as danceathon, swimathon,<br />
etc... (Marchand 1969: 213, emphases orig.)<br />
Whether Marchand’s rather extreme claim is justi<strong>fi</strong>ed or not, he may well<br />
consider this battle as lost. What was considered “a purely dictionary interest”<br />
(see Marchand 1969: 132) a few decades ago, is now an integral part <strong>of</strong> the<br />
English lexicon without doubt. <strong>The</strong> main results <strong>of</strong> this study indicate that what<br />
has here been called combining forms are indeed frequent in English, and they<br />
combine quite freely with both native and non-native elements.<br />
This study has attempted to measure the <strong>productivity</strong> <strong>of</strong> six<br />
neoclassical combining forms (three initial and three <strong>fi</strong>nal) in English, using<br />
several complementary methods. First, a quantitative, statistical measure P<br />
developed by Baayen and his co-workers has been applied to these combining<br />
forms. Secondly, two measures <strong>of</strong> lexical statistics, i.e., vocabulary growth curves<br />
and frequency spectra, have been used in order to take as many aspects <strong>of</strong><br />
<strong>productivity</strong> into account as possible. <strong>The</strong> motivation for using these methods has<br />
been discussed, as well as the various problems that have occurred during the<br />
research process and the appropriate solutions.<br />
As regards their P values, the initial combining forms seemed to<br />
score somewhat higher than the <strong>fi</strong>nal combining forms. In terms <strong>of</strong> vocabulary<br />
growth curves, the initial combining forms produced steeper curves than the <strong>fi</strong>nal<br />
combining forms, which indicates that their vocabulary grows at a greater pace<br />
65
and that they are therefore more productive. <strong>The</strong> third measure, i.e., the<br />
comparison <strong>of</strong> observed and expected frequency spectra, did not yield any major<br />
differences between the initial and the <strong>fi</strong>nal combining forms. <strong>The</strong> observed and<br />
the expected values lined up rather nicely with almost all <strong>of</strong> the <strong>selected</strong> elements.<br />
<strong>The</strong> <strong>fi</strong>ndings are not surprising, given the latest tendencies in<br />
English word-formation that seem to favour pre<strong>fi</strong>x-like elements over suf<strong>fi</strong>xes. It<br />
has even been suggested that a typological shift might be under way in the English<br />
word-formation system (Bauer 2003a). In addition, the proliferation <strong>of</strong> scienti<strong>fi</strong>c<br />
texts seems to contribute to the high <strong>productivity</strong> <strong>of</strong> these elements. In order to<br />
make broader generalizations, though, a larger set <strong>of</strong> combining forms should be<br />
compared. However, the problem is that in performing an analysis that draws<br />
from lexical statistics, sample size matters: many combining forms do not produce<br />
many hits even in a large corpus like the BNC. This makes their statistical analysis<br />
hard. R might fail to produce expected frequency spectra for some combining<br />
forms if the sample size is too small. For example, the initial combining form<br />
ultra- has 2000 occurrences in the BNC, which is not enough for R to produce an<br />
extrapolated vocabulary growth curve.<br />
It must also be noted that most <strong>of</strong> the combining forms <strong>selected</strong><br />
turned out to be rather af<strong>fi</strong>x-like by nature, their function being quantifying rather<br />
than that <strong>of</strong> resembling a lexeme. This is probably related to their high frequency<br />
in English, as opposed to more lexeme-like combining forms. However, these<br />
combining forms with more clearly lexeme-like meanings (such as bio- ‘life’ or<br />
astro- ‘star’) would probably have acquired different, probably lower, P values.<br />
Another important aspect that must be kept in mind at this point is<br />
psycholinguistics. Whether speakers actually differentiate between af<strong>fi</strong>xes and<br />
combining forms from the point <strong>of</strong> view <strong>of</strong> mental lexicon and cognitive<br />
processing is a question worth investigating. As Plag has noted, “speakers <strong>of</strong><br />
English can and do master English morphology without etymological knowledge”<br />
(2002: 286). It is the boundary between pre<strong>fi</strong>xes and initial combining forms that<br />
seems to be rather blurry. Final combining forms, on the other hand, seem to be<br />
more distinct from af<strong>fi</strong>xes as a category.<br />
During the research process, it has also become evident that<br />
Baayen’s quantitative method measures only one aspect <strong>of</strong> <strong>productivity</strong>, i.e., the<br />
growth rate <strong>of</strong> the vocabulary <strong>of</strong> a certain <strong>morphological</strong> category (see Baayen<br />
66
2008: 902). As Plag has put it, it is “only a more or less accurate statement <strong>of</strong> the<br />
problem and not a solution to it” (1999: 35). <strong>The</strong>refore the potential <strong>productivity</strong><br />
P cannot be the only way to assess the <strong>productivity</strong> <strong>of</strong> the <strong>selected</strong> pre<strong>fi</strong>xes and<br />
combining forms. In this study, methods <strong>of</strong> lexical statistics were used to <strong>of</strong>fer an<br />
account that is as diverse as possible and that takes into account the different<br />
aspects <strong>of</strong> <strong>morphological</strong> <strong>productivity</strong>.<br />
Another interesting topic for future research could involve<br />
comparing the <strong>productivity</strong> rates <strong>of</strong> combining forms across different registers,<br />
following the example <strong>of</strong> Plag et al. (1999). Many Greek pre<strong>fi</strong>xes in particular are<br />
relatively recent loans in English and are mainly used in word-formation in<br />
specialized scienti<strong>fi</strong>c registers (Biber et al. 1999: 324). <strong>The</strong> problem with that kind<br />
<strong>of</strong> research is that the samples compared must be <strong>of</strong> equal size, which, as Säily<br />
has noted, may lead to discarding valuable data (2011: 125). On the other hand,<br />
studying the correlation <strong>of</strong> <strong>productivity</strong> and different sociolinguistic variables,<br />
such as gender, age or social rank <strong>of</strong> the speakers (see Säily and Suomela 2009,<br />
Säily 2011), is also worth further study. Another aspect <strong>of</strong> <strong>productivity</strong> that has<br />
not been thoroughly investigated yet is the competition between different wordformation<br />
processes that express the same need to name various concepts and<br />
phenomena (see Lüdeling et al. 2006).<br />
In the future, more theoretical accounts <strong>of</strong> <strong>productivity</strong> are also<br />
needed: lately, studies <strong>of</strong> <strong>productivity</strong> have mainly concentrated on the<br />
<strong>productivity</strong> <strong>of</strong> individual word-formation processes (see Dressler and Ladányi<br />
2000: 104). Besides, even though the development <strong>of</strong> different empirical and<br />
statistical methods to measure <strong>productivity</strong> has flourished in the past few years,<br />
there are still many questions that concern the very nature <strong>of</strong> <strong>productivity</strong> that<br />
need to be tackled. Plag, for example, mentions such problems as allomorphy and<br />
the relationship between rival <strong>morphological</strong> processes, as well as how<br />
morphology interacts with phonology, syntax, and semantics (1999: 1). In<br />
addition, the study <strong>of</strong> <strong>productivity</strong> might have practical applications beyond<br />
morphology, measuring the “generative potential” (Evert/Baroni) <strong>of</strong> such things<br />
as syntactic processes or collocations/idioms. LNRE models in particular are also<br />
an area that has not so far been widely employed in the study <strong>of</strong> <strong>productivity</strong>.<br />
Kastovsky argues that instead <strong>of</strong> considering the demarcation<br />
between combining forms and af<strong>fi</strong>xation, a more relevant question would actually<br />
67
e the difference between compounding and af<strong>fi</strong>xation in general (2009: 10–11).<br />
He points out that a number <strong>of</strong> native af<strong>fi</strong>xes has developed from members <strong>of</strong><br />
compounds (e.g., -dom, -hood, and -wise), and the same phenomenon can be seen<br />
with combining forms (2009: 10). Kastovsky suggests that the theoretical<br />
framework <strong>of</strong> grammaticalization would probably be productive in this kind <strong>of</strong><br />
analysis (2009: 11). From the semantic point <strong>of</strong> view, a crucial distinction in this<br />
model is also the one between “lexically speci<strong>fi</strong>c” and “lexically non-speci<strong>fi</strong>c”<br />
word-formation processes, which can also be seen as a cline between words and<br />
stems on the one hand and af<strong>fi</strong>xes on the other hand (Kastovsky 2009: 11).<br />
Another research topic not yet fully covered is the so-called modern combining<br />
forms, such as jazzo-(phile) or speedo-(meter) (Prćić 2005, 2007, 2008).<br />
Primary sources and s<strong>of</strong>tware<br />
BNC = <strong>The</strong> British National Corpus, version 4.1 (BNC XML Edition). 2008.<br />
Distributed by Oxford University Computing Services on behalf <strong>of</strong> the<br />
BNC Consortium. Online: http://www.natcorp.ox.ac.uk/<br />
COCA = <strong>The</strong> corpus <strong>of</strong> contemporary American English. Compiled by M. Davies.<br />
2008– Online: http://corpus.byu.edu/coca<br />
<strong>The</strong> R project for statistical computing. http://www.r-project.org<br />
References<br />
Amiot, D. and G. Dal. 2007. Integrating neoclassical combining forms into a<br />
lexeme-based morphology. In G. Booij.et al. (eds.), On-line proceedings<br />
<strong>of</strong> the <strong>fi</strong>fth Mediterranean Morphology Meeting (MMM5) Fréjus 15–18<br />
September 2005, University <strong>of</strong> Bologna. Online:<br />
http://mmm.lingue.unibo.it/mmm-proc/MMM5/232-336-Amiot-Dal.pdf<br />
[accessed on October 26, 2012]<br />
68
Anshen, F. and M. Aron<strong>of</strong>f. 1981. Morphological <strong>productivity</strong> and phonological<br />
transparency. <strong>The</strong> Canadian Journal <strong>of</strong> Linguistics 26 (1), 63–72.<br />
Aron<strong>of</strong>f, M. 1976. Word formation in generative grammar. Cambridge, MA: MIT<br />
Press.<br />
Aron<strong>of</strong>f, M. and F. Anshen. 1998. Morphology and the lexicon: Lexicalization<br />
and <strong>productivity</strong>. In A. Spencer and A.M. Zwicky (eds.), <strong>The</strong> Handbook<br />
<strong>of</strong> morphology. (Blackwell Handbooks in Linguistics). Cambridge, MA:<br />
Blackwell Publishers, 237–248.<br />
Baayen, R. H. 1992. Quantitative aspects <strong>of</strong> <strong>morphological</strong> <strong>productivity</strong>. In G.<br />
Booij and J. van Marle (eds.), Yearbook <strong>of</strong> morphology 1991.<br />
Dordrecht: Kluwer, 109–149.<br />
Baayen, R. H. 1993. On frequency, transparency and <strong>productivity</strong>. In G. Booij and<br />
J. van Marle (eds.), Yearbook <strong>of</strong> morphology 1992. Dordrecht: Kluwer,<br />
182–208.<br />
Baayen, R.H. 2001. Word frequency distributions. Dordrecht: Kluwer.<br />
Baayen, R.H. 2003. Probabilistic approaches to morphology. In R. Bod et al.<br />
(eds.), Probabilistic linguistics. Cambridge, MA, MIT Press, 229–287.<br />
Baayen, R. H. 2008. Analysing linguistic data: a practical introduction to<br />
statistics using R. Cambridge: Cambridge University Press.<br />
Baayen, R. H. 2009. Corpus linguistics in morphology: <strong>morphological</strong><br />
<strong>productivity</strong>. In A. Lüdeling and M. Kytö (eds.), Corpus linguistics: An<br />
international handbook, ed. by Anke Lüdeling and Merja Kytö. Berlin:<br />
Walter de Gruyter, 899–919.<br />
Baayen, R. H. and R. Lieber. 1991. Productivity and English derivation: A<br />
corpus-based study. Linguistics 29 (5), 801–843.<br />
Baayen, R. H. and A. Renouf. 1996. Chronicling the Times: Lexical innovations<br />
in an English newspaper. Language 72 (1), 69–96.<br />
Barker, C. 1998. Episodic -ee in English: a thematic role constraint on new word<br />
formation. Language 74: 695–727.<br />
Baroni, M. 2008. Distributions in text. In A. Lüdeling and M. Kytö (eds.), Corpus<br />
linguistics: an international handbook. Berlin: Mouton de Gruyter,<br />
803–822.<br />
Baroni, M. and S. Evert. 2008. Statistical methods for corpus exploitation. In A.<br />
Lüdeling and M. Kytö (eds.), Corpus linguistics: an international<br />
handbook. Berlin: Mouton de Gruyter, 777–803.<br />
69
Bauer, L. 1983. English word-formation. (Cambridge Textbooks in Linguistics).<br />
Cambridge: Cambridge University Press.<br />
Bauer, L. 1988. Introducing linguistic morphology. Edinburgh: Edinburgh<br />
University Press.<br />
Bauer, L. 1992. Scalar <strong>productivity</strong> and -lily adverbs. In G. Booij and J. van Marle<br />
(eds.), Yearbook <strong>of</strong> morphology 1991. Dordrecht: Kluwer, 185–191.<br />
Bauer. L. 1995. Is <strong>morphological</strong> <strong>productivity</strong> non-linguistic? Acta Linguistica<br />
Hungarica 43 (1–2), 19–31.<br />
Bauer, L. 1998. Is there a class <strong>of</strong> neoclassical compounds and if so is it<br />
productive? Linguistics 36 (3), 403–422.<br />
Bauer, L.. 2001. Morphological <strong>productivity</strong>. (Cambridge Studies in Linguistics).<br />
Cambridge: Cambridge University Press.<br />
Bauer, L. 2003a. English pre<strong>fi</strong>xation – a typological shift? Acta Linguistica<br />
Hungarica 50 (1–2), 33–40.<br />
Bauer, L. 2003b. Introducing linguistic morphology. 2nd edition. Edinburgh:<br />
Edinburgh University Press.<br />
Bauer, L. 2003c.<strong>The</strong> <strong>productivity</strong> <strong>of</strong> (non-)productive morphology. Rivista di<br />
Linguistica 15, 7–16.<br />
Bauer, L. 2005 Productivity: theories. In Štekauer and R. Lieber, 315–334.<br />
Biber, D. et al. 1999. Longman grammar <strong>of</strong> spoken and written English. London:<br />
Longman.<br />
Bloch, B. and G.L. Trager. 1942. Outline <strong>of</strong> linguistic analysis. Baltimore:<br />
Linguistic Society <strong>of</strong> America.<br />
Booij, G. E. 2005. <strong>The</strong> grammar <strong>of</strong> words: an introduction to linguistic<br />
morphology. Oxford: Oxford University Press.<br />
Burnard, L. (ed.). 2007. Reference guide for the British National Corpus (XML<br />
edition). Published for the British National Corpus Consortium by the<br />
Research Technologies Service at Oxford University Computing<br />
Services. Online: http://www.natcorp.ox.ac.uk/docs/URG/ (accessed on<br />
1 October 2012).<br />
Burnley, L. 1992. Lexis and semantics. In Blake, N. (ed.) <strong>The</strong> Cambridge history<br />
<strong>of</strong> the English language, Vol. 2: 1066–1476. 409–499.<br />
Carstairs-McCarthy, A. 1992. Current morphology. London/New York:<br />
Routledge.<br />
70
Carstairs-McCarthy, A. 2002. An introduction to English morphology: words and<br />
their structure. (Edinburgh Textbooks on the English Language).<br />
Edinburgh: Edinburgh University Press.<br />
CEEC = Corpus <strong>of</strong> Early English Correspondence. 1998. Compiled by Terttu<br />
Nevalainen, Helena Raumolin-Brunberg, Jukka Keränen, Minna<br />
Nevala, Arja Nurmi and Minna Palander-Collin at the Department <strong>of</strong><br />
English, University <strong>of</strong> <strong>Helsinki</strong>.<br />
Chitashvili, R.J. and H. Baayen 1993. Word frequency distributions. In L.<br />
Hrebícek and G. Altmann (eds.) Quantitative text analysis. Trier :<br />
WVT, 54-135.<br />
Corbin, D. 1987. Morphologie dérivationnelle et structuration du lexique.<br />
Tübingen: Max Niemeyer.<br />
Cottez, H. 1985. Dictionnaire des structures du vocabulaire savant: éléments et<br />
modèles de formation. Paris: Le Robert.<br />
Cowie, C. 1998. <strong>The</strong> discourse motivations for neologising: Action<br />
nominatization in the history <strong>of</strong> English. In J. Coleman and C. J. Kay<br />
(eds.), Lexicology, semantics and lexicography. Selected papers from<br />
the fourth G. L. Brook Symposium.(Amsterdam Studies in the <strong>The</strong>ory<br />
and History <strong>of</strong> Linguistic Science. Series 4, Current issues in linguistic<br />
theory, vol. 194). Amsterdam: John Benjamins.179–207.<br />
Cowie, C. 2003. “Uncommon terminations”: Proscription and <strong>morphological</strong><br />
<strong>productivity</strong>. Rivista di Linguistica 15 (1), 17–30.<br />
Cowie, C. and C. Dalton-Puffer. 2002. Diachronic word-formation and studying<br />
changes in <strong>productivity</strong> over time: <strong>The</strong>oretical and methodological<br />
considerations. In J. E. Díaz Vera (ed.) A changing world <strong>of</strong> words:<br />
Studies in English historical lexicography, lexicology and semantics.<br />
Amsterdam: Rodopi.410–437.<br />
Dalton-Puffer, C. 1994.Productive or not productive? <strong>The</strong> Romance element in<br />
Middle English derivation. in F. Férnandez et al. (eds.) English<br />
Historical Linguistics 1992: Papers from the 7th International<br />
Conference on English Historical Linguistics, Valencia, 22–26<br />
September 1992. Amsterdam: John Benjamins.247–260.<br />
Dalton-Puffer, C. 1996.<strong>The</strong> French influence on Middle English morphology: A<br />
corpus-based study <strong>of</strong> derivation. (Topics in English Linguistics 20).<br />
Berlin: Mouton de Gruyter.<br />
71
Dressler, W. U. and M. Ladányi. 2000. Productivity in word formation (WF): A<br />
<strong>morphological</strong> approach. Acta Linguistica Hungarica 47 (1–4), 103–<br />
144.<br />
Evert, S. and A. Lüdeling. 2001. Measuring <strong>morphological</strong> <strong>productivity</strong>: Is<br />
automatic preprocessing suf<strong>fi</strong>cient? In P. Rayson et al. (eds.)<br />
Proceedings <strong>of</strong> the Corpus Linguistics 2001 Conference. Lancaster:<br />
University Center for Computer Research on Language. 167–<br />
175.Online:<br />
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.28.9162&rep<br />
=rep1&type=pdf [accessed on October 15, 2012]<br />
Evert, S. and A. Lüdeling. 2003. Linguistic experience and <strong>productivity</strong>: Corpus<br />
evidence for <strong>fi</strong>ne-grained distinctions. In D. Archer et al. (eds.)<br />
Proceedings <strong>of</strong> the Corpus Linguistics 2003 Conference. UCREL<br />
technical paper 16. Lancaster: UCREL. 475-483.Online:<br />
http://citeseer.ist.psu.edu/viewdoc/similar?doi=10.1.1.13.2813&type=ab<br />
[accessed on January 13, 2013]<br />
Evert, S. and M. Baroni. 2007. zipfR: Word frequency distributions in R.<br />
Proceedings <strong>of</strong> the 45 th Annual Meeting <strong>of</strong> the Association <strong>of</strong><br />
Computational Linguistics, Posters and Demonstrations Session.<br />
Prague, Czech Republic. Online: http://www.stefanevert.de/PUB/EvertBaroni2007.pdf<br />
[retrieved on April 13, 2013]<br />
Férnandez-Domínguez, J. et al. 2007. How is low <strong>morphological</strong> <strong>productivity</strong><br />
measured? Atlantis 29 (1), 29–54.<br />
Férnandez-Domínguez, J. 2009.Productivity in English word-formation: An<br />
approach to n+n compounding. (European University Studies, series<br />
21, Linguistics, vol. 341) New York: Peter Lang.<br />
Fischer, R. 1998. Lexical change in present-day English: A corpus-based study <strong>of</strong><br />
the motivation, institutionalization, and <strong>productivity</strong> <strong>of</strong> creative<br />
neologisms. (Language in Performance). Tübingen: G. Narr.<br />
Fradin, B. 2000. Combining forms, blends and related phenomena. In U.<br />
Doleschal and A. M. Thornton (eds.), Extragrammatical and<br />
marginal morphology. Munich: LINCOM Europa, 11–60.<br />
Frauenfelder, U. H. and R. Schreuder 1992.Constraining psycholinguistic models<br />
<strong>of</strong> <strong>morphological</strong> processing and representation: the role <strong>of</strong><br />
<strong>productivity</strong>. In G. E. Booij and J. van Marle (eds.), Yearbook <strong>of</strong><br />
morphology 1991. Dordrecht: Kluwer Academic Publishers, 165–183.<br />
Harastani, R. et al. 2012. Neoclassical compound alignments from comparable<br />
corpora. In A. Gelbukh (ed.), Computational linguistics and intelligent<br />
72
text processing 13 th international conference, CICLing 2012, New<br />
Delhi, India, March 11–17, 2012, Proceedings, Part II. Heidelberg:<br />
Springer, 72–82.<br />
Haspelmath, M. 2002. Understanding morphology.(Understanding Language<br />
Series). London: Arnold.<br />
Hay, J. 2001. Lexical frequency in morphology: Is everything relative?<br />
Linguistics 39, 1041–1070.<br />
Hay, J. 2003. Causes and consequences <strong>of</strong> word structure. (Outstanding<br />
Dissertations in Linguistics). New York: Routledge.<br />
Hay, J. and H. Baayen. 2002. Parsing and <strong>productivity</strong>. In G. Booij and J. van<br />
Marle (eds.), Yearbook <strong>of</strong> morphology 2002, 203-235. Dordrecht:<br />
Kluwer.<br />
Hohenhaus, P. 2005. Where word-formation cannot extend the vocabulary:<br />
Creativity, <strong>productivity</strong> and the lexicon in synchronic and diachronic<br />
morphology. In D. Kastovsky and A. Mettinger (eds.), Lexical change<br />
and the genesis <strong>of</strong> English vocabulary. Berlin/New York: Mouton de<br />
Gruyter,<br />
Huddleston, R. and G. Pullum. 2002. <strong>The</strong> Cambridge grammar <strong>of</strong> the English<br />
language. Cambridge: Cambridge University Press.<br />
Hughes, G. 2003. A history <strong>of</strong> English words. Oxford: Blackwell Publishing.<br />
Kastovsky, D. 2001. English morphology: A typological reappraisal. In C.<br />
Schaner-Wolles et al. (eds.). Naturally! Linguistic studies in honour <strong>of</strong><br />
Wolfgang Ulrich Dressler presented on the occasion <strong>of</strong> his 60th<br />
birthday. Turin: Rosenberg and Sellier.215–224.<br />
Kastovsky, D. 2006. Typological changes in derivational morphology. In A. van<br />
Kemenade and B. Los (eds.), <strong>The</strong> handbook <strong>of</strong> the history <strong>of</strong> English.<br />
(Blackwell Handbooks in Linguistics). Malden: Blackwell Publishing.<br />
151–176<br />
Kastovsky, D. 2009. Astronaut, astrology, astrophysics: About combining forms,<br />
classical compounds and af<strong>fi</strong>xoids. In R. W. McConchie et al. (eds.)<br />
Selected proceedings <strong>of</strong> the 2008 Symposium on New Approaches in<br />
English Historical Lexis (HEL-LEX 2). Somerville, MA: Cascadilla<br />
Proceedings project. 1-13. Online: http://www.lingref.com/cpp/hellex/2008/paper2161.pdf<br />
[accessed on January 14, 2013].<br />
Krott, A. et al. 1999. Complex words in complex worlds. Linguistics 37. 905–926.<br />
73
Lee, D. 2001. Genres, registers, text types, domains, and styles: clarifying the<br />
concepts and navigating a path through the BNC jungle. Language<br />
Learning & Technology 5 (3), 37–72.<br />
Lehrer, A. 1998.Scapes, holics, and thons: <strong>The</strong> semantics <strong>of</strong> English combining<br />
forms. American Speech 73 (1), 3–28.<br />
Lieber, Rochelle. 2005. English word-formation processes: Observations, issues,<br />
and thoughts on future research. In Štekauer and Lieber (eds.), 375–427.<br />
Lipka, L. 1990. An outline <strong>of</strong> English lexicology. (Forschung & Studium Anglistik<br />
3). Tübingen: Niemeyer.<br />
Lüdeling, A. et al. 2000. On measuring <strong>morphological</strong> <strong>productivity</strong>. Proceedings<br />
<strong>of</strong> the KONVENS 2000, 57–61. Online:<br />
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.106.4819&re<br />
p=rep1&type=pdf [accessed on November 16th, 2012].<br />
Lüdeling, A. et al. 2002. Neoclassical word-formation in German. In Booij.G. and<br />
J. van Marle (eds.) Yearbook <strong>of</strong> morphology 2001.Dordrecht: Kluwer,<br />
253–283.<br />
Lüdeling, A. and S. Evert. 2003. Linguistic experience and <strong>productivity</strong>: corpus<br />
evidence for <strong>fi</strong>ne-grained distinctions. Proceedings <strong>of</strong> the Corpus<br />
Linguistics 2003 Conference. Online: http://www.ims.unistuttgart.de/projekte/corplex/paper/evert/LuedelingEvert2003.pdf<br />
[accessed on November 26, 2012].<br />
Lüdeling, A. et al. 2006. Need and competition: Deconstructing quantitative<br />
<strong>productivity</strong>. Online: http://cogsci.uniosnabrueck.de/~qitl/abstracts/luedeling_baroni_evert.pdf<br />
[accessed on<br />
November 25, 2012].<br />
Marchand, H. 1969. <strong>The</strong> categories and types <strong>of</strong> present-day English wordformation:<br />
A synchronic-diachronic approach. 2nd edition. Munich:<br />
Beck’sche.<br />
Marle, J. van. 1985. On the paradigmatic dimension <strong>of</strong> <strong>morphological</strong> creativity.<br />
(Publications in language sciences 15). Dordrecht: Foris Publications.<br />
McCauley, J. 2006. Technical combining forms in the third edition <strong>of</strong> the OED:<br />
Word-formation in a historical dictionary. In R. W. McConchie et al.<br />
(eds.) Selected proceedings <strong>of</strong> the 2005 Symposium on New Approaches<br />
in English Historical Lexis (HEL-LEX).Somerville, MA: Cascadilla<br />
Proceedings Project. 95–104. Online: http://www.lingref.com/cpp/hellex/2005/paper1350.pdf<br />
[accessed on January 14, 2013]<br />
74
McConchie, R. W. 1998. <strong>The</strong> vernacularization <strong>of</strong> the negative pre<strong>fi</strong>x dis- in Early<br />
Modern English. Lexicology, semantics and lexicography. In J.<br />
Coleman and C. J. Kay (eds.) Selected papers from the fourth G. L.<br />
Brook Symposium.(Amsterdam Studies in the <strong>The</strong>ory and History <strong>of</strong><br />
Linguistic Science. Series 4, Current issues in linguistic theory, vol.<br />
194) Amsterdam: John Benjamins. 209–227.<br />
McConchie, R. W. 2006. Disseisin: <strong>The</strong> lexeme and the legal fact in Early Middle<br />
English. In R. Dury el al. (eds.) English historical linguistics 2006:<br />
Lexical and semantic change. Amsterdam: John Benjamins. 203–216.<br />
Minkova, D. and R. Stockwell. 2009. English words: Structure and history. 2nd<br />
edition. Cambridge: Cambridge University Press.<br />
Miller, D.G. 2012. External influences on English: From its beginnings to the<br />
renaissance. Oxford: Oxford University Press.<br />
Nevalainen, T. 1999. Early Modern English lexis and semantics. In Lass, R. (ed.).<br />
<strong>The</strong> Cambridge history <strong>of</strong> the English language, Vol. 3: 1476–1776.<br />
Cambridge: Cambridge University Press. 332–458.<br />
OED = <strong>The</strong> Oxford English dictionary online.2007 (Online version 2013).3rd<br />
edition. Oxford: Oxford University Press.Online: http://www.oed.com<br />
Palmer, C. C. 2008. Borrowed derivational morphology in Late Middle English: a<br />
study <strong>of</strong> the records <strong>of</strong> the London Grocers and Goldsmiths. In<br />
Fitzmaurice, S. et al. (eds.). Studies in the history <strong>of</strong> the English<br />
language IV: empirical and analytical advances in the study <strong>of</strong> English<br />
language change. (Topics in English linguistics 61). Berlin/New York:<br />
Mouton de Gruyter, 231-64.<br />
Palmer, C. C. 2009. Borrowings, derivational morphology, and perceived<br />
<strong>productivity</strong> in English, 1300–1600.Ph. D. thesis. University <strong>of</strong><br />
Michigan. Online:<br />
http://deepblue.lib.umich.edu/bitstream/2027.42/64624/1/palmercc_1.pd<br />
f [retrieved September 18, 2012]<br />
Pápai, V. 2004. Explicitation: A universal <strong>of</strong> translated text? In A. Mauranen and<br />
P. Kujamäki (eds.) Translation universals: Do they exist? Amsterdam:<br />
John Benjamins Publishing.143-164.<br />
Plag, I. 1999. Morphological <strong>productivity</strong>. Structural constraints in English<br />
derivation.(Topics in English Linguistics 28). Berlin: Mouton de<br />
Gruyter.<br />
75
Plag, I. 2002. <strong>The</strong> role <strong>of</strong> selectional restrictions, phonotactics, and parsing in<br />
constraining suf<strong>fi</strong>x ordering in English. In G. Booij and J. van Marle<br />
(eds.), Yearbook <strong>of</strong> morphology 2001, 285–314.Dordrecht: Kluwer.<br />
Plag, I. 2003, Word-formation in English. (Cambridge Textbooks in Linguistics).<br />
Cambridge: Cambridge University Press.<br />
Plag, I. 2006. Productivity. In B. Aarts and A. McMahon (eds.), <strong>The</strong> handbook <strong>of</strong><br />
English linguistics. (Blackwell Handbooks in Linguistics). Malden:<br />
Blackwell Publishing.537–556.<br />
Plag, I. et al. 1999. Morphological <strong>productivity</strong> across speech and writing. English<br />
Language and Linguistics 3 (2), 209–228.<br />
Pounder, A. 2000. Processes and paradigms in word-formation morphology.<br />
Berlin: Mouton de Gruyter.<br />
Prćić, T. 2005. Pre<strong>fi</strong>xes vs. initial combining forms in English: A lexicographic<br />
perspective. International Journal <strong>of</strong> Lexicography 18 (3), 313–334.<br />
Prćić, T. 2007. Headhood <strong>of</strong> suf<strong>fi</strong>xes and <strong>fi</strong>nal combining forms in English wordformation.<br />
Acta linguistic hungarica 54 (4), 381–392.<br />
Prćić, T. 2008. Suf<strong>fi</strong>xes vs. <strong>fi</strong>nal combining forms in English: A lexicographic<br />
perspective. International Journal <strong>of</strong> Lexicography 21 (1), 1–22.<br />
Quirk, R. et al. 1985. A comprehensive grammar <strong>of</strong> the English language.<br />
London: Longman.<br />
Rainer, F. 2005. Constraints on <strong>productivity</strong>. In Štekauer and Lieber (eds.), 335-<br />
352.<br />
Renouf, Antoinette. 2007. Tracing lexical <strong>productivity</strong> and creativity in the British<br />
Media: ‘<strong>The</strong> Chavs and the Chav-nots’. In J. Munat (ed.) Lexical<br />
creativity. texts and contexts. (Studies in Functional and Structural<br />
Linguistics). Amsterdam: Benjamins. 61–89.<br />
Säily, T. 2011. Variation in <strong>morphological</strong> <strong>productivity</strong> in the BNC:<br />
Sociolinguistic and methodological considerations. Corpus Linguistics<br />
and Linguistic <strong>The</strong>ory 7 (1), 119–141.<br />
Säily, T. and J. Suomela. 2009. Comparing type counts: <strong>The</strong> case <strong>of</strong> women, men<br />
and -ity in early English letters. In A. Renouf and A. Kehoe (eds.)<br />
Corpus linguistics: Re<strong>fi</strong>nements and reassessments. (Language and<br />
Computers: Studies in Practical Linguistics 69.) Amsterdam: Rodopi.<br />
871–909. Authors’ version available from<br />
http://www.cs.helsinki.<strong>fi</strong>/u/josuomel/doc/icame-2007-paper.pdf<br />
[retrieved on October 27, 2012]<br />
76
Schäfer, J. 1989. Early Modern English: OED, New OED, EMED. In R. W.<br />
Bailey (ed.), Dictionaries <strong>of</strong> English: prospects for the record <strong>of</strong> our<br />
language. Cambridge: Cambridge University Press. 66-74.<br />
Štekauer, P. and R. Lieber (eds.) 2005.Handbook <strong>of</strong> word-formation. Dordrecht:<br />
Springer.<br />
Štekauer, P. et al. 2005.Word-formation as creativity within <strong>productivity</strong><br />
constraints: sociolinguistic evidence. Onomasiology Online 6, 1–55.<br />
Online:<br />
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.92.9286&rep<br />
=rep1&type=pdf [Accessed on November 18, 2012]<br />
Szymanek, B. 2005. <strong>The</strong> latest trends in English word-formation. In Štekauer and<br />
Lieber (eds.), 429–448.<br />
Tweedie, F. J. and R. H. Baayen. 1998. How variable may a constant be?<br />
Measures <strong>of</strong> lexical richness in perspective. Computers and the<br />
Humanities 32, 323–352.<br />
Warren, B. 1990.<strong>The</strong> importance <strong>of</strong> combining forms. In W. U. Dressler et al<br />
(eds.) Contemporary morphology.Trends in linguistics: Studies and<br />
monographs 49. Berlin: Mouton de Gruyter.111–132.<br />
Zipf, G. K. 1949. Human behaviour and the principle <strong>of</strong> least effort. Cambridge,<br />
MA: Addison-Wesley.<br />
Zipf, G. K. 1965. <strong>The</strong> psycho-biology <strong>of</strong> language. Cambridge, MA: MIT Press.<br />
77